From 8cadd2831bf3abc94f4530e7fdbab7bb39b6b27d Mon Sep 17 00:00:00 2001 From: Jesse Barnes Date: Mon, 10 May 2010 14:26:20 -0700 Subject: timer: add on-stack deferrable timer interfaces In some cases (for instance with kernel threads) it may be desireable to use on-stack deferrable timers to get their power saving benefits. Add interfaces to support this for the IPS driver. Signed-off-by: Jesse Barnes Signed-off-by: Matthew Garrett --- include/linux/timer.h | 15 +++++++++++++++ 1 file changed, 15 insertions(+) (limited to 'include') diff --git a/include/linux/timer.h b/include/linux/timer.h index ea965b857a50..38cf093ef62c 100644 --- a/include/linux/timer.h +++ b/include/linux/timer.h @@ -100,6 +100,13 @@ void init_timer_deferrable_key(struct timer_list *timer, setup_timer_on_stack_key((timer), #timer, &__key, \ (fn), (data)); \ } while (0) +#define setup_deferrable_timer_on_stack(timer, fn, data) \ + do { \ + static struct lock_class_key __key; \ + setup_deferrable_timer_on_stack_key((timer), #timer, \ + &__key, (fn), \ + (data)); \ + } while (0) #else #define init_timer(timer)\ init_timer_key((timer), NULL, NULL) @@ -111,6 +118,8 @@ void init_timer_deferrable_key(struct timer_list *timer, setup_timer_key((timer), NULL, NULL, (fn), (data)) #define setup_timer_on_stack(timer, fn, data)\ setup_timer_on_stack_key((timer), NULL, NULL, (fn), (data)) +#define setup_deferrable_timer_on_stack(timer, fn, data)\ + setup_deferrable_timer_on_stack_key((timer), NULL, NULL, (fn), (data)) #endif #ifdef CONFIG_DEBUG_OBJECTS_TIMERS @@ -150,6 +159,12 @@ static inline void setup_timer_on_stack_key(struct timer_list *timer, init_timer_on_stack_key(timer, name, key); } +extern void setup_deferrable_timer_on_stack_key(struct timer_list *timer, + const char *name, + struct lock_class_key *key, + void (*function)(unsigned long), + unsigned long data); + /** * timer_pending - is a timer pending? * @timer: the timer in question -- cgit v1.2.3-70-g09d2 From aa7ffc01d254c91a36bf854d57a14049c6134c72 Mon Sep 17 00:00:00 2001 From: Jesse Barnes Date: Fri, 14 May 2010 15:41:14 -0700 Subject: x86 platform driver: intelligent power sharing driver Intel Core i3/5 platforms with integrated graphics support both CPU and GPU turbo mode. CPU turbo mode is opportunistic: the CPU will use any available power to increase core frequencies if thermal headroom is available. The GPU side is more manual however; the graphics driver must monitor GPU power and temperature and coordinate with a core thermal driver to take advantage of available thermal and power headroom in the package. The intelligent power sharing (IPS) driver is intended to coordinate this activity by monitoring MCP (multi-chip package) temperature and power, allowing the CPU and/or GPU to increase their power consumption, and thus performance, when possible. The goal is to maximize performance within a given platform's TDP (thermal design point). Signed-off-by: Jesse Barnes Signed-off-by: Matthew Garrett --- drivers/platform/x86/Kconfig | 10 + drivers/platform/x86/Makefile | 1 + drivers/platform/x86/intel_ips.c | 1655 ++++++++++++++++++++++++++++++++++++++ include/drm/i915_drm.h | 9 + 4 files changed, 1675 insertions(+) create mode 100644 drivers/platform/x86/intel_ips.c (limited to 'include') diff --git a/drivers/platform/x86/Kconfig b/drivers/platform/x86/Kconfig index fd060016b7e9..724b2ed1a3cb 100644 --- a/drivers/platform/x86/Kconfig +++ b/drivers/platform/x86/Kconfig @@ -539,4 +539,14 @@ config INTEL_SCU_IPC some embedded Intel x86 platforms. This is not needed for PC-type machines. +config INTEL_IPS + tristate "Intel Intelligent Power Sharing" + depends on ACPI + ---help--- + Intel Calpella platforms support dynamic power sharing between the + CPU and GPU, maximizing performance in a given TDP. This driver, + along with the CPU frequency and i915 drivers, provides that + functionality. If in doubt, say Y here; it will only load on + supported platforms. + endif # X86_PLATFORM_DEVICES diff --git a/drivers/platform/x86/Makefile b/drivers/platform/x86/Makefile index 8770bfe71431..7318fc2c1629 100644 --- a/drivers/platform/x86/Makefile +++ b/drivers/platform/x86/Makefile @@ -26,3 +26,4 @@ obj-$(CONFIG_TOPSTAR_LAPTOP) += topstar-laptop.o obj-$(CONFIG_ACPI_TOSHIBA) += toshiba_acpi.o obj-$(CONFIG_TOSHIBA_BT_RFKILL) += toshiba_bluetooth.o obj-$(CONFIG_INTEL_SCU_IPC) += intel_scu_ipc.o +obj-$(CONFIG_INTEL_IPS) += intel_ips.o diff --git a/drivers/platform/x86/intel_ips.c b/drivers/platform/x86/intel_ips.c new file mode 100644 index 000000000000..f1dce3b8372d --- /dev/null +++ b/drivers/platform/x86/intel_ips.c @@ -0,0 +1,1655 @@ +/* + * Copyright (c) 2009-2010 Intel Corporation + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * The full GNU General Public License is included in this distribution in + * the file called "COPYING". + * + * Authors: + * Jesse Barnes + */ + +/* + * Some Intel Ibex Peak based platforms support so-called "intelligent + * power sharing", which allows the CPU and GPU to cooperate to maximize + * performance within a given TDP (thermal design point). This driver + * performs the coordination between the CPU and GPU, monitors thermal and + * power statistics in the platform, and initializes power monitoring + * hardware. It also provides a few tunables to control behavior. Its + * primary purpose is to safely allow CPU and GPU turbo modes to be enabled + * by tracking power and thermal budget; secondarily it can boost turbo + * performance by allocating more power or thermal budget to the CPU or GPU + * based on available headroom and activity. + * + * The basic algorithm is driven by a 5s moving average of tempurature. If + * thermal headroom is available, the CPU and/or GPU power clamps may be + * adjusted upwards. If we hit the thermal ceiling or a thermal trigger, + * we scale back the clamp. Aside from trigger events (when we're critically + * close or over our TDP) we don't adjust the clamps more than once every + * five seconds. + * + * The thermal device (device 31, function 6) has a set of registers that + * are updated by the ME firmware. The ME should also take the clamp values + * written to those registers and write them to the CPU, but we currently + * bypass that functionality and write the CPU MSR directly. + * + * UNSUPPORTED: + * - dual MCP configs + * + * TODO: + * - handle CPU hotplug + * - provide turbo enable/disable api + * - make sure we can write turbo enable/disable reg based on MISC_EN + * + * Related documents: + * - CDI 403777, 403778 - Auburndale EDS vol 1 & 2 + * - CDI 401376 - Ibex Peak EDS + * - ref 26037, 26641 - IPS BIOS spec + * - ref 26489 - Nehalem BIOS writer's guide + * - ref 26921 - Ibex Peak BIOS Specification + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define PCI_DEVICE_ID_INTEL_THERMAL_SENSOR 0x3b32 + +/* + * Package level MSRs for monitor/control + */ +#define PLATFORM_INFO 0xce +#define PLATFORM_TDP (1<<29) +#define PLATFORM_RATIO (1<<28) + +#define IA32_MISC_ENABLE 0x1a0 +#define IA32_MISC_TURBO_EN (1ULL<<38) + +#define TURBO_POWER_CURRENT_LIMIT 0x1ac +#define TURBO_TDC_OVR_EN (1UL<<31) +#define TURBO_TDC_MASK (0x000000007fff0000UL) +#define TURBO_TDC_SHIFT (16) +#define TURBO_TDP_OVR_EN (1UL<<15) +#define TURBO_TDP_MASK (0x0000000000003fffUL) + +/* + * Core/thread MSRs for monitoring + */ +#define IA32_PERF_CTL 0x199 +#define IA32_PERF_TURBO_DIS (1ULL<<32) + +/* + * Thermal PCI device regs + */ +#define THM_CFG_TBAR 0x10 +#define THM_CFG_TBAR_HI 0x14 + +#define THM_TSIU 0x00 +#define THM_TSE 0x01 +#define TSE_EN 0xb8 +#define THM_TSS 0x02 +#define THM_TSTR 0x03 +#define THM_TSTTP 0x04 +#define THM_TSCO 0x08 +#define THM_TSES 0x0c +#define THM_TSGPEN 0x0d +#define TSGPEN_HOT_LOHI (1<<1) +#define TSGPEN_CRIT_LOHI (1<<2) +#define THM_TSPC 0x0e +#define THM_PPEC 0x10 +#define THM_CTA 0x12 +#define THM_PTA 0x14 +#define PTA_SLOPE_MASK (0xff00) +#define PTA_SLOPE_SHIFT 8 +#define PTA_OFFSET_MASK (0x00ff) +#define THM_MGTA 0x16 +#define MGTA_SLOPE_MASK (0xff00) +#define MGTA_SLOPE_SHIFT 8 +#define MGTA_OFFSET_MASK (0x00ff) +#define THM_TRC 0x1a +#define TRC_CORE2_EN (1<<15) +#define TRC_THM_EN (1<<12) +#define TRC_C6_WAR (1<<8) +#define TRC_CORE1_EN (1<<7) +#define TRC_CORE_PWR (1<<6) +#define TRC_PCH_EN (1<<5) +#define TRC_MCH_EN (1<<4) +#define TRC_DIMM4 (1<<3) +#define TRC_DIMM3 (1<<2) +#define TRC_DIMM2 (1<<1) +#define TRC_DIMM1 (1<<0) +#define THM_TES 0x20 +#define THM_TEN 0x21 +#define TEN_UPDATE_EN 1 +#define THM_PSC 0x24 +#define PSC_NTG (1<<0) /* No GFX turbo support */ +#define PSC_NTPC (1<<1) /* No CPU turbo support */ +#define PSC_PP_DEF (0<<2) /* Perf policy up to driver */ +#define PSP_PP_PC (1<<2) /* BIOS prefers CPU perf */ +#define PSP_PP_BAL (2<<2) /* BIOS wants balanced perf */ +#define PSP_PP_GFX (3<<2) /* BIOS prefers GFX perf */ +#define PSP_PBRT (1<<4) /* BIOS run time support */ +#define THM_CTV1 0x30 +#define CTV_TEMP_ERROR (1<<15) +#define CTV_TEMP_MASK 0x3f +#define CTV_ +#define THM_CTV2 0x32 +#define THM_CEC 0x34 /* undocumented power accumulator in joules */ +#define THM_AE 0x3f +#define THM_HTS 0x50 /* 32 bits */ +#define HTS_PCPL_MASK (0x7fe00000) +#define HTS_PCPL_SHIFT 21 +#define HTS_GPL_MASK (0x001ff000) +#define HTS_GPL_SHIFT 12 +#define HTS_PP_MASK (0x00000c00) +#define HTS_PP_SHIFT 10 +#define HTS_PP_DEF 0 +#define HTS_PP_PROC 1 +#define HTS_PP_BAL 2 +#define HTS_PP_GFX 3 +#define HTS_PCTD_DIS (1<<9) +#define HTS_GTD_DIS (1<<8) +#define HTS_PTL_MASK (0x000000fe) +#define HTS_PTL_SHIFT 1 +#define HTS_NVV (1<<0) +#define THM_HTSHI 0x54 /* 16 bits */ +#define HTS2_PPL_MASK (0x03ff) +#define HTS2_PRST_MASK (0x3c00) +#define HTS2_PRST_SHIFT 10 +#define HTS2_PRST_UNLOADED 0 +#define HTS2_PRST_RUNNING 1 +#define HTS2_PRST_TDISOP 2 /* turbo disabled due to power */ +#define HTS2_PRST_TDISHT 3 /* turbo disabled due to high temp */ +#define HTS2_PRST_TDISUSR 4 /* user disabled turbo */ +#define HTS2_PRST_TDISPLAT 5 /* platform disabled turbo */ +#define HTS2_PRST_TDISPM 6 /* power management disabled turbo */ +#define HTS2_PRST_TDISERR 7 /* some kind of error disabled turbo */ +#define THM_PTL 0x56 +#define THM_MGTV 0x58 +#define TV_MASK 0x000000000000ff00 +#define TV_SHIFT 8 +#define THM_PTV 0x60 +#define PTV_MASK 0x00ff +#define THM_MMGPC 0x64 +#define THM_MPPC 0x66 +#define THM_MPCPC 0x68 +#define THM_TSPIEN 0x82 +#define TSPIEN_AUX_LOHI (1<<0) +#define TSPIEN_HOT_LOHI (1<<1) +#define TSPIEN_CRIT_LOHI (1<<2) +#define TSPIEN_AUX2_LOHI (1<<3) +#define THM_TSLOCK 0x83 +#define THM_ATR 0x84 +#define THM_TOF 0x87 +#define THM_STS 0x98 +#define STS_PCPL_MASK (0x7fe00000) +#define STS_PCPL_SHIFT 21 +#define STS_GPL_MASK (0x001ff000) +#define STS_GPL_SHIFT 12 +#define STS_PP_MASK (0x00000c00) +#define STS_PP_SHIFT 10 +#define STS_PP_DEF 0 +#define STS_PP_PROC 1 +#define STS_PP_BAL 2 +#define STS_PP_GFX 3 +#define STS_PCTD_DIS (1<<9) +#define STS_GTD_DIS (1<<8) +#define STS_PTL_MASK (0x000000fe) +#define STS_PTL_SHIFT 1 +#define STS_NVV (1<<0) +#define THM_SEC 0x9c +#define SEC_ACK (1<<0) +#define THM_TC3 0xa4 +#define THM_TC1 0xa8 +#define STS_PPL_MASK (0x0003ff00) +#define STS_PPL_SHIFT 16 +#define THM_TC2 0xac +#define THM_DTV 0xb0 +#define THM_ITV 0xd8 +#define ITV_ME_SEQNO_MASK 0x000f0000 /* ME should update every ~200ms */ +#define ITV_ME_SEQNO_SHIFT (16) +#define ITV_MCH_TEMP_MASK 0x0000ff00 +#define ITV_MCH_TEMP_SHIFT (8) +#define ITV_PCH_TEMP_MASK 0x000000ff + +#define thm_readb(off) readb(ips->regmap + (off)) +#define thm_readw(off) readw(ips->regmap + (off)) +#define thm_readl(off) readl(ips->regmap + (off)) +#define thm_readq(off) readq(ips->regmap + (off)) + +#define thm_writeb(off, val) writeb((val), ips->regmap + (off)) +#define thm_writew(off, val) writew((val), ips->regmap + (off)) +#define thm_writel(off, val) writel((val), ips->regmap + (off)) + +static const int IPS_ADJUST_PERIOD = 5000; /* ms */ + +/* For initial average collection */ +static const int IPS_SAMPLE_PERIOD = 200; /* ms */ +static const int IPS_SAMPLE_WINDOW = 5000; /* 5s moving window of samples */ +#define IPS_SAMPLE_COUNT (IPS_SAMPLE_WINDOW / IPS_SAMPLE_PERIOD) + +/* Per-SKU limits */ +struct ips_mcp_limits { + int cpu_family; + int cpu_model; /* includes extended model... */ + int mcp_power_limit; /* mW units */ + int core_power_limit; + int mch_power_limit; + int core_temp_limit; /* degrees C */ + int mch_temp_limit; +}; + +/* Max temps are -10 degrees C to avoid PROCHOT# */ + +struct ips_mcp_limits ips_sv_limits = { + .mcp_power_limit = 35000, + .core_power_limit = 29000, + .mch_power_limit = 20000, + .core_temp_limit = 95, + .mch_temp_limit = 90 +}; + +struct ips_mcp_limits ips_lv_limits = { + .mcp_power_limit = 25000, + .core_power_limit = 21000, + .mch_power_limit = 13000, + .core_temp_limit = 95, + .mch_temp_limit = 90 +}; + +struct ips_mcp_limits ips_ulv_limits = { + .mcp_power_limit = 18000, + .core_power_limit = 14000, + .mch_power_limit = 11000, + .core_temp_limit = 95, + .mch_temp_limit = 90 +}; + +struct ips_driver { + struct pci_dev *dev; + void *regmap; + struct task_struct *monitor; + struct task_struct *adjust; + struct dentry *debug_root; + + /* Average CPU core temps (all averages in .01 degrees C for precision) */ + u16 ctv1_avg_temp; + u16 ctv2_avg_temp; + /* GMCH average */ + u16 mch_avg_temp; + /* Average for the CPU (both cores?) */ + u16 mcp_avg_temp; + /* Average power consumption (in mW) */ + u32 cpu_avg_power; + u32 mch_avg_power; + + /* Offset values */ + u16 cta_val; + u16 pta_val; + u16 mgta_val; + + /* Maximums & prefs, protected by turbo status lock */ + spinlock_t turbo_status_lock; + u16 mcp_temp_limit; + u16 mcp_power_limit; + u16 core_power_limit; + u16 mch_power_limit; + bool cpu_turbo_enabled; + bool __cpu_turbo_on; + bool gpu_turbo_enabled; + bool __gpu_turbo_on; + bool gpu_preferred; + bool poll_turbo_status; + bool second_cpu; + struct ips_mcp_limits *limits; + + /* Optional MCH interfaces for if i915 is in use */ + unsigned long (*read_mch_val)(void); + bool (*gpu_raise)(void); + bool (*gpu_lower)(void); + bool (*gpu_busy)(void); + bool (*gpu_turbo_disable)(void); + + /* For restoration at unload */ + u64 orig_turbo_limit; + u64 orig_turbo_ratios; +}; + +/** + * ips_cpu_busy - is CPU busy? + * @ips: IPS driver struct + * + * Check CPU for load to see whether we should increase its thermal budget. + * + * RETURNS: + * True if the CPU could use more power, false otherwise. + */ +static bool ips_cpu_busy(struct ips_driver *ips) +{ + if ((avenrun[0] >> FSHIFT) > 1) + return true; + + return false; +} + +/** + * ips_cpu_raise - raise CPU power clamp + * @ips: IPS driver struct + * + * Raise the CPU power clamp by %IPS_CPU_STEP, in accordance with TDP for + * this platform. + * + * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR upwards (as + * long as we haven't hit the TDP limit for the SKU). + */ +static void ips_cpu_raise(struct ips_driver *ips) +{ + u64 turbo_override; + u16 cur_tdp_limit, new_tdp_limit; + + if (!ips->cpu_turbo_enabled) + return; + + rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override); + + cur_tdp_limit = turbo_override & TURBO_TDP_MASK; + new_tdp_limit = cur_tdp_limit + 8; /* 1W increase */ + + /* Clamp to SKU TDP limit */ + if (((new_tdp_limit * 10) / 8) > ips->core_power_limit) + new_tdp_limit = cur_tdp_limit; + + thm_writew(THM_MPCPC, (new_tdp_limit * 10) / 8); + + turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDC_OVR_EN; + wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override); + + turbo_override &= ~TURBO_TDP_MASK; + turbo_override |= new_tdp_limit; + + wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override); +} + +/** + * ips_cpu_lower - lower CPU power clamp + * @ips: IPS driver struct + * + * Lower CPU power clamp b %IPS_CPU_STEP if possible. + * + * We do this by adjusting the TURBO_POWER_CURRENT_LIMIT MSR down, going + * as low as the platform limits will allow (though we could go lower there + * wouldn't be much point). + */ +static void ips_cpu_lower(struct ips_driver *ips) +{ + u64 turbo_override; + u16 cur_limit, new_limit; + + rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override); + + cur_limit = turbo_override & TURBO_TDP_MASK; + new_limit = cur_limit - 8; /* 1W decrease */ + + /* Clamp to SKU TDP limit */ + if (((new_limit * 10) / 8) < (ips->orig_turbo_limit & TURBO_TDP_MASK)) + new_limit = ips->orig_turbo_limit & TURBO_TDP_MASK; + + thm_writew(THM_MPCPC, (new_limit * 10) / 8); + + turbo_override |= TURBO_TDC_OVR_EN | TURBO_TDC_OVR_EN; + wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override); + + turbo_override &= ~TURBO_TDP_MASK; + turbo_override |= new_limit; + + wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override); +} + +/** + * do_enable_cpu_turbo - internal turbo enable function + * @data: unused + * + * Internal function for actually updating MSRs. When we enable/disable + * turbo, we need to do it on each CPU; this function is the one called + * by on_each_cpu() when needed. + */ +static void do_enable_cpu_turbo(void *data) +{ + u64 perf_ctl; + + rdmsrl(IA32_PERF_CTL, perf_ctl); + if (perf_ctl & IA32_PERF_TURBO_DIS) { + perf_ctl &= ~IA32_PERF_TURBO_DIS; + wrmsrl(IA32_PERF_CTL, perf_ctl); + } +} + +/** + * ips_enable_cpu_turbo - enable turbo mode on all CPUs + * @ips: IPS driver struct + * + * Enable turbo mode by clearing the disable bit in IA32_PERF_CTL on + * all logical threads. + */ +static void ips_enable_cpu_turbo(struct ips_driver *ips) +{ + /* Already on, no need to mess with MSRs */ + if (ips->__cpu_turbo_on) + return; + + on_each_cpu(do_enable_cpu_turbo, ips, 1); + + ips->__cpu_turbo_on = true; +} + +/** + * do_disable_cpu_turbo - internal turbo disable function + * @data: unused + * + * Internal function for actually updating MSRs. When we enable/disable + * turbo, we need to do it on each CPU; this function is the one called + * by on_each_cpu() when needed. + */ +static void do_disable_cpu_turbo(void *data) +{ + u64 perf_ctl; + + rdmsrl(IA32_PERF_CTL, perf_ctl); + if (!(perf_ctl & IA32_PERF_TURBO_DIS)) { + perf_ctl |= IA32_PERF_TURBO_DIS; + wrmsrl(IA32_PERF_CTL, perf_ctl); + } +} + +/** + * ips_disable_cpu_turbo - disable turbo mode on all CPUs + * @ips: IPS driver struct + * + * Disable turbo mode by setting the disable bit in IA32_PERF_CTL on + * all logical threads. + */ +static void ips_disable_cpu_turbo(struct ips_driver *ips) +{ + /* Already off, leave it */ + if (!ips->__cpu_turbo_on) + return; + + on_each_cpu(do_disable_cpu_turbo, ips, 1); + + ips->__cpu_turbo_on = false; +} + +/** + * ips_gpu_busy - is GPU busy? + * @ips: IPS driver struct + * + * Check GPU for load to see whether we should increase its thermal budget. + * We need to call into the i915 driver in this case. + * + * RETURNS: + * True if the GPU could use more power, false otherwise. + */ +static bool ips_gpu_busy(struct ips_driver *ips) +{ + return false; +} + +/** + * ips_gpu_raise - raise GPU power clamp + * @ips: IPS driver struct + * + * Raise the GPU frequency/power if possible. We need to call into the + * i915 driver in this case. + */ +static void ips_gpu_raise(struct ips_driver *ips) +{ + if (!ips->gpu_turbo_enabled) + return; + + if (!ips->gpu_raise()) + ips->gpu_turbo_enabled = false; + + return; +} + +/** + * ips_gpu_lower - lower GPU power clamp + * @ips: IPS driver struct + * + * Lower GPU frequency/power if possible. Need to call i915. + */ +static void ips_gpu_lower(struct ips_driver *ips) +{ + if (!ips->gpu_turbo_enabled) + return; + + if (!ips->gpu_lower()) + ips->gpu_turbo_enabled = false; + + return; +} + +/** + * ips_enable_gpu_turbo - notify the gfx driver turbo is available + * @ips: IPS driver struct + * + * Call into the graphics driver indicating that it can safely use + * turbo mode. + */ +static void ips_enable_gpu_turbo(struct ips_driver *ips) +{ + if (ips->__gpu_turbo_on) + return; + ips->__gpu_turbo_on = true; +} + +/** + * ips_disable_gpu_turbo - notify the gfx driver to disable turbo mode + * @ips: IPS driver struct + * + * Request that the graphics driver disable turbo mode. + */ +static void ips_disable_gpu_turbo(struct ips_driver *ips) +{ + /* Avoid calling i915 if turbo is already disabled */ + if (!ips->__gpu_turbo_on) + return; + + if (!ips->gpu_turbo_disable()) + dev_err(&ips->dev->dev, "failed to disable graphis turbo\n"); + else + ips->__gpu_turbo_on = false; +} + +/** + * mcp_exceeded - check whether we're outside our thermal & power limits + * @ips: IPS driver struct + * + * Check whether the MCP is over its thermal or power budget. + */ +static bool mcp_exceeded(struct ips_driver *ips) +{ + unsigned long flags; + bool ret = false; + + spin_lock_irqsave(&ips->turbo_status_lock, flags); + if (ips->mcp_avg_temp > (ips->mcp_temp_limit * 100)) + ret = true; + if (ips->cpu_avg_power + ips->mch_avg_power > ips->mcp_power_limit) + ret = true; + spin_unlock_irqrestore(&ips->turbo_status_lock, flags); + + if (ret) + dev_warn(&ips->dev->dev, + "MCP power or thermal limit exceeded\n"); + + return ret; +} + +/** + * cpu_exceeded - check whether a CPU core is outside its limits + * @ips: IPS driver struct + * @cpu: CPU number to check + * + * Check a given CPU's average temp or power is over its limit. + */ +static bool cpu_exceeded(struct ips_driver *ips, int cpu) +{ + unsigned long flags; + int avg; + bool ret = false; + + spin_lock_irqsave(&ips->turbo_status_lock, flags); + avg = cpu ? ips->ctv2_avg_temp : ips->ctv1_avg_temp; + if (avg > (ips->limits->core_temp_limit * 100)) + ret = true; + if (ips->cpu_avg_power > ips->core_power_limit) + ret = true; + spin_unlock_irqrestore(&ips->turbo_status_lock, flags); + + if (ret) + dev_warn(&ips->dev->dev, + "CPU power or thermal limit exceeded\n"); + + return ret; +} + +/** + * mch_exceeded - check whether the GPU is over budget + * @ips: IPS driver struct + * + * Check the MCH temp & power against their maximums. + */ +static bool mch_exceeded(struct ips_driver *ips) +{ + unsigned long flags; + bool ret = false; + + spin_lock_irqsave(&ips->turbo_status_lock, flags); + if (ips->mch_avg_temp > (ips->limits->mch_temp_limit * 100)) + ret = true; + spin_unlock_irqrestore(&ips->turbo_status_lock, flags); + + return ret; +} + +/** + * update_turbo_limits - get various limits & settings from regs + * @ips: IPS driver struct + * + * Update the IPS power & temp limits, along with turbo enable flags, + * based on latest register contents. + * + * Used at init time and for runtime BIOS support, which requires polling + * the regs for updates (as a result of AC->DC transition for example). + * + * LOCKING: + * Caller must hold turbo_status_lock (outside of init) + */ +static void update_turbo_limits(struct ips_driver *ips) +{ + u32 hts = thm_readl(THM_HTS); + + ips->cpu_turbo_enabled = !(hts & HTS_PCTD_DIS); + ips->gpu_turbo_enabled = !(hts & HTS_GTD_DIS); + ips->core_power_limit = thm_readw(THM_MPCPC); + ips->mch_power_limit = thm_readw(THM_MMGPC); + ips->mcp_temp_limit = thm_readw(THM_PTL); + ips->mcp_power_limit = thm_readw(THM_MPPC); + + /* Ignore BIOS CPU vs GPU pref */ +} + +/** + * ips_adjust - adjust power clamp based on thermal state + * @data: ips driver structure + * + * Wake up every 5s or so and check whether we should adjust the power clamp. + * Check CPU and GPU load to determine which needs adjustment. There are + * several things to consider here: + * - do we need to adjust up or down? + * - is CPU busy? + * - is GPU busy? + * - is CPU in turbo? + * - is GPU in turbo? + * - is CPU or GPU preferred? (CPU is default) + * + * So, given the above, we do the following: + * - up (TDP available) + * - CPU not busy, GPU not busy - nothing + * - CPU busy, GPU not busy - adjust CPU up + * - CPU not busy, GPU busy - adjust GPU up + * - CPU busy, GPU busy - adjust preferred unit up, taking headroom from + * non-preferred unit if necessary + * - down (at TDP limit) + * - adjust both CPU and GPU down if possible + * + cpu+ gpu+ cpu+gpu- cpu-gpu+ cpu-gpu- +cpu < gpu < cpu+gpu+ cpu+ gpu+ nothing +cpu < gpu >= cpu+gpu-(mcp<) cpu+gpu-(mcp<) gpu- gpu- +cpu >= gpu < cpu-gpu+(mcp<) cpu- cpu-gpu+(mcp<) cpu- +cpu >= gpu >= cpu-gpu- cpu-gpu- cpu-gpu- cpu-gpu- + * + */ +static int ips_adjust(void *data) +{ + struct ips_driver *ips = data; + unsigned long flags; + + dev_dbg(&ips->dev->dev, "starting ips-adjust thread\n"); + + /* + * Adjust CPU and GPU clamps every 5s if needed. Doing it more + * often isn't recommended due to ME interaction. + */ + do { + bool cpu_busy = ips_cpu_busy(ips); + bool gpu_busy = ips_gpu_busy(ips); + + spin_lock_irqsave(&ips->turbo_status_lock, flags); + if (ips->poll_turbo_status) + update_turbo_limits(ips); + spin_unlock_irqrestore(&ips->turbo_status_lock, flags); + + /* Update turbo status if necessary */ + if (ips->cpu_turbo_enabled) + ips_enable_cpu_turbo(ips); + else + ips_disable_cpu_turbo(ips); + + if (ips->gpu_turbo_enabled) + ips_enable_gpu_turbo(ips); + else + ips_disable_gpu_turbo(ips); + + /* We're outside our comfort zone, crank them down */ + if (!mcp_exceeded(ips)) { + ips_cpu_lower(ips); + ips_gpu_lower(ips); + goto sleep; + } + + if (!cpu_exceeded(ips, 0) && cpu_busy) + ips_cpu_raise(ips); + else + ips_cpu_lower(ips); + + if (!mch_exceeded(ips) && gpu_busy) + ips_gpu_raise(ips); + else + ips_gpu_lower(ips); + +sleep: + schedule_timeout_interruptible(msecs_to_jiffies(IPS_ADJUST_PERIOD)); + } while (!kthread_should_stop()); + + dev_dbg(&ips->dev->dev, "ips-adjust thread stopped\n"); + + return 0; +} + +/* + * Helpers for reading out temp/power values and calculating their + * averages for the decision making and monitoring functions. + */ + +static u16 calc_avg_temp(struct ips_driver *ips, u16 *array) +{ + u64 total = 0; + int i; + u16 avg; + + for (i = 0; i < IPS_SAMPLE_COUNT; i++) + total += (u64)(array[i] * 100); + + do_div(total, IPS_SAMPLE_COUNT); + + avg = (u16)total; + + return avg; +} + +static u16 read_mgtv(struct ips_driver *ips) +{ + u16 ret; + u64 slope, offset; + u64 val; + + val = thm_readq(THM_MGTV); + val = (val & TV_MASK) >> TV_SHIFT; + + slope = offset = thm_readw(THM_MGTA); + slope = (slope & MGTA_SLOPE_MASK) >> MGTA_SLOPE_SHIFT; + offset = offset & MGTA_OFFSET_MASK; + + ret = ((val * slope + 0x40) >> 7) + offset; + + + return ret; +} + +static u16 read_ptv(struct ips_driver *ips) +{ + u16 val, slope, offset; + + slope = (ips->pta_val & PTA_SLOPE_MASK) >> PTA_SLOPE_SHIFT; + offset = ips->pta_val & PTA_OFFSET_MASK; + + val = thm_readw(THM_PTV) & PTV_MASK; + + return val; +} + +static u16 read_ctv(struct ips_driver *ips, int cpu) +{ + int reg = cpu ? THM_CTV2 : THM_CTV1; + u16 val; + + val = thm_readw(reg); + if (!(val & CTV_TEMP_ERROR)) + val = (val) >> 6; /* discard fractional component */ + else + val = 0; + + return val; +} + +static u32 get_cpu_power(struct ips_driver *ips, u32 *last, int period) +{ + u32 val; + u32 ret; + + /* + * CEC is in joules/65535. Take difference over time to + * get watts. + */ + val = thm_readl(THM_CEC); + + /* period is in ms and we want mW */ + ret = (((val - *last) * 1000) / period); + ret = (ret * 1000) / 65535; + *last = val; + + return ret; +} + +static const u16 temp_decay_factor = 2; +static u16 update_average_temp(u16 avg, u16 val) +{ + u16 ret; + + /* Multiply by 100 for extra precision */ + ret = (val * 100 / temp_decay_factor) + + (((temp_decay_factor - 1) * avg) / temp_decay_factor); + return ret; +} + +static const u16 power_decay_factor = 2; +static u16 update_average_power(u32 avg, u32 val) +{ + u32 ret; + + ret = (val / power_decay_factor) + + (((power_decay_factor - 1) * avg) / power_decay_factor); + + return ret; +} + +static u32 calc_avg_power(struct ips_driver *ips, u32 *array) +{ + u64 total = 0; + u32 avg; + int i; + + for (i = 0; i < IPS_SAMPLE_COUNT; i++) + total += array[i]; + + do_div(total, IPS_SAMPLE_COUNT); + avg = (u32)total; + + return avg; +} + +static void monitor_timeout(unsigned long arg) +{ + wake_up_process((struct task_struct *)arg); +} + +/** + * ips_monitor - temp/power monitoring thread + * @data: ips driver structure + * + * This is the main function for the IPS driver. It monitors power and + * tempurature in the MCP and adjusts CPU and GPU power clams accordingly. + * + * We keep a 5s moving average of power consumption and tempurature. Using + * that data, along with CPU vs GPU preference, we adjust the power clamps + * up or down. + */ +static int ips_monitor(void *data) +{ + struct ips_driver *ips = data; + struct timer_list timer; + unsigned long seqno_timestamp, expire, last_msecs, last_sample_period; + int i; + u32 *cpu_samples = NULL, *mchp_samples = NULL, old_cpu_power; + u16 *mcp_samples = NULL, *ctv1_samples = NULL, *ctv2_samples = NULL, + *mch_samples = NULL; + u8 cur_seqno, last_seqno; + + mcp_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL); + ctv1_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL); + ctv2_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL); + mch_samples = kzalloc(sizeof(u16) * IPS_SAMPLE_COUNT, GFP_KERNEL); + cpu_samples = kzalloc(sizeof(u32) * IPS_SAMPLE_COUNT, GFP_KERNEL); + mchp_samples = kzalloc(sizeof(u32) * IPS_SAMPLE_COUNT, GFP_KERNEL); + if (!mcp_samples || !ctv1_samples || !ctv2_samples || !mch_samples) { + dev_err(&ips->dev->dev, + "failed to allocate sample array, ips disabled\n"); + kfree(mcp_samples); + kfree(ctv1_samples); + kfree(ctv2_samples); + kfree(mch_samples); + kfree(cpu_samples); + kthread_stop(ips->adjust); + return -ENOMEM; + } + + last_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >> + ITV_ME_SEQNO_SHIFT; + seqno_timestamp = get_jiffies_64(); + + old_cpu_power = thm_readl(THM_CEC) / 65535; + schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD)); + + /* Collect an initial average */ + for (i = 0; i < IPS_SAMPLE_COUNT; i++) { + u32 mchp, cpu_power; + u16 val; + + mcp_samples[i] = read_ptv(ips); + + val = read_ctv(ips, 0); + ctv1_samples[i] = val; + + val = read_ctv(ips, 1); + ctv2_samples[i] = val; + + val = read_mgtv(ips); + mch_samples[i] = val; + + cpu_power = get_cpu_power(ips, &old_cpu_power, + IPS_SAMPLE_PERIOD); + cpu_samples[i] = cpu_power; + + if (ips->read_mch_val) { + mchp = ips->read_mch_val(); + mchp_samples[i] = mchp; + } + + schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD)); + if (kthread_should_stop()) + break; + } + + ips->mcp_avg_temp = calc_avg_temp(ips, mcp_samples); + ips->ctv1_avg_temp = calc_avg_temp(ips, ctv1_samples); + ips->ctv2_avg_temp = calc_avg_temp(ips, ctv2_samples); + ips->mch_avg_temp = calc_avg_temp(ips, mch_samples); + ips->cpu_avg_power = calc_avg_power(ips, cpu_samples); + ips->mch_avg_power = calc_avg_power(ips, mchp_samples); + kfree(mcp_samples); + kfree(ctv1_samples); + kfree(ctv2_samples); + kfree(mch_samples); + kfree(cpu_samples); + kfree(mchp_samples); + + /* Start the adjustment thread now that we have data */ + wake_up_process(ips->adjust); + + /* + * Ok, now we have an initial avg. From here on out, we track the + * running avg using a decaying average calculation. This allows + * us to reduce the sample frequency if the CPU and GPU are idle. + */ + old_cpu_power = thm_readl(THM_CEC); + schedule_timeout_interruptible(msecs_to_jiffies(IPS_SAMPLE_PERIOD)); + last_sample_period = IPS_SAMPLE_PERIOD; + + setup_deferrable_timer_on_stack(&timer, monitor_timeout, + (unsigned long)current); + do { + u32 cpu_val, mch_val; + u16 val; + + /* MCP itself */ + val = read_ptv(ips); + ips->mcp_avg_temp = update_average_temp(ips->mcp_avg_temp, val); + + /* Processor 0 */ + val = read_ctv(ips, 0); + ips->ctv1_avg_temp = + update_average_temp(ips->ctv1_avg_temp, val); + /* Power */ + cpu_val = get_cpu_power(ips, &old_cpu_power, + last_sample_period); + ips->cpu_avg_power = + update_average_power(ips->cpu_avg_power, cpu_val); + + if (ips->second_cpu) { + /* Processor 1 */ + val = read_ctv(ips, 1); + ips->ctv2_avg_temp = + update_average_temp(ips->ctv2_avg_temp, val); + } + + /* MCH */ + val = read_mgtv(ips); + ips->mch_avg_temp = update_average_temp(ips->mch_avg_temp, val); + /* Power */ + if (ips->read_mch_val) { + mch_val = ips->read_mch_val(); + ips->mch_avg_power = + update_average_power(ips->mch_avg_power, + mch_val); + } + + /* + * Make sure ME is updating thermal regs. + * Note: + * If it's been more than a second since the last update, + * the ME is probably hung. + */ + cur_seqno = (thm_readl(THM_ITV) & ITV_ME_SEQNO_MASK) >> + ITV_ME_SEQNO_SHIFT; + if (cur_seqno == last_seqno && + time_after(jiffies, seqno_timestamp + HZ)) { + dev_warn(&ips->dev->dev, "ME failed to update for more than 1s, likely hung\n"); + } else { + seqno_timestamp = get_jiffies_64(); + last_seqno = cur_seqno; + } + + last_msecs = jiffies_to_msecs(jiffies); + expire = jiffies + msecs_to_jiffies(IPS_SAMPLE_PERIOD); + + __set_current_state(TASK_UNINTERRUPTIBLE); + mod_timer(&timer, expire); + schedule(); + + /* Calculate actual sample period for power averaging */ + last_sample_period = jiffies_to_msecs(jiffies) - last_msecs; + if (!last_sample_period) + last_sample_period = 1; + } while (!kthread_should_stop()); + + del_timer_sync(&timer); + destroy_timer_on_stack(&timer); + + dev_dbg(&ips->dev->dev, "ips-monitor thread stopped\n"); + + return 0; +} + +#if 0 +#define THM_DUMPW(reg) \ + { \ + u16 val = thm_readw(reg); \ + dev_dbg(&ips->dev->dev, #reg ": 0x%04x\n", val); \ + } +#define THM_DUMPL(reg) \ + { \ + u32 val = thm_readl(reg); \ + dev_dbg(&ips->dev->dev, #reg ": 0x%08x\n", val); \ + } +#define THM_DUMPQ(reg) \ + { \ + u64 val = thm_readq(reg); \ + dev_dbg(&ips->dev->dev, #reg ": 0x%016x\n", val); \ + } + +static void dump_thermal_info(struct ips_driver *ips) +{ + u16 ptl; + + ptl = thm_readw(THM_PTL); + dev_dbg(&ips->dev->dev, "Processor temp limit: %d\n", ptl); + + THM_DUMPW(THM_CTA); + THM_DUMPW(THM_TRC); + THM_DUMPW(THM_CTV1); + THM_DUMPL(THM_STS); + THM_DUMPW(THM_PTV); + THM_DUMPQ(THM_MGTV); +} +#endif + +/** + * ips_irq_handler - handle temperature triggers and other IPS events + * @irq: irq number + * @arg: unused + * + * Handle temperature limit trigger events, generally by lowering the clamps. + * If we're at a critical limit, we clamp back to the lowest possible value + * to prevent emergency shutdown. + */ +static irqreturn_t ips_irq_handler(int irq, void *arg) +{ + struct ips_driver *ips = arg; + u8 tses = thm_readb(THM_TSES); + u8 tes = thm_readb(THM_TES); + + if (!tses && !tes) + return IRQ_NONE; + + dev_info(&ips->dev->dev, "TSES: 0x%02x\n", tses); + dev_info(&ips->dev->dev, "TES: 0x%02x\n", tes); + + /* STS update from EC? */ + if (tes & 1) { + u32 sts, tc1; + + sts = thm_readl(THM_STS); + tc1 = thm_readl(THM_TC1); + + if (sts & STS_NVV) { + spin_lock(&ips->turbo_status_lock); + ips->core_power_limit = (sts & STS_PCPL_MASK) >> + STS_PCPL_SHIFT; + ips->mch_power_limit = (sts & STS_GPL_MASK) >> + STS_GPL_SHIFT; + /* ignore EC CPU vs GPU pref */ + ips->cpu_turbo_enabled = !(sts & STS_PCTD_DIS); + ips->gpu_turbo_enabled = !(sts & STS_GTD_DIS); + ips->mcp_temp_limit = (sts & STS_PTL_MASK) >> + STS_PTL_SHIFT; + ips->mcp_power_limit = (tc1 & STS_PPL_MASK) >> + STS_PPL_SHIFT; + spin_unlock(&ips->turbo_status_lock); + + thm_writeb(THM_SEC, SEC_ACK); + } + thm_writeb(THM_TES, tes); + } + + /* Thermal trip */ + if (tses) { + dev_warn(&ips->dev->dev, + "thermal trip occurred, tses: 0x%04x\n", tses); + thm_writeb(THM_TSES, tses); + } + + return IRQ_HANDLED; +} + +#ifndef CONFIG_DEBUG_FS +static void ips_debugfs_init(struct ips_driver *ips) { return; } +static void ips_debugfs_cleanup(struct ips_driver *ips) { return; } +#else + +/* Expose current state and limits in debugfs if possible */ + +struct ips_debugfs_node { + struct ips_driver *ips; + char *name; + int (*show)(struct seq_file *m, void *data); +}; + +static int show_cpu_temp(struct seq_file *m, void *data) +{ + struct ips_driver *ips = m->private; + + seq_printf(m, "%d.%02d\n", ips->ctv1_avg_temp / 100, + ips->ctv1_avg_temp % 100); + + return 0; +} + +static int show_cpu_power(struct seq_file *m, void *data) +{ + struct ips_driver *ips = m->private; + + seq_printf(m, "%dmW\n", ips->cpu_avg_power); + + return 0; +} + +static int show_cpu_clamp(struct seq_file *m, void *data) +{ + u64 turbo_override; + int tdp, tdc; + + rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override); + + tdp = (int)(turbo_override & TURBO_TDP_MASK); + tdc = (int)((turbo_override & TURBO_TDC_MASK) >> TURBO_TDC_SHIFT); + + /* Convert to .1W/A units */ + tdp = tdp * 10 / 8; + tdc = tdc * 10 / 8; + + /* Watts Amperes */ + seq_printf(m, "%d.%dW %d.%dA\n", tdp / 10, tdp % 10, + tdc / 10, tdc % 10); + + return 0; +} + +static int show_mch_temp(struct seq_file *m, void *data) +{ + struct ips_driver *ips = m->private; + + seq_printf(m, "%d.%02d\n", ips->mch_avg_temp / 100, + ips->mch_avg_temp % 100); + + return 0; +} + +static int show_mch_power(struct seq_file *m, void *data) +{ + struct ips_driver *ips = m->private; + + seq_printf(m, "%dmW\n", ips->mch_avg_power); + + return 0; +} + +static struct ips_debugfs_node ips_debug_files[] = { + { NULL, "cpu_temp", show_cpu_temp }, + { NULL, "cpu_power", show_cpu_power }, + { NULL, "cpu_clamp", show_cpu_clamp }, + { NULL, "mch_temp", show_mch_temp }, + { NULL, "mch_power", show_mch_power }, +}; + +static int ips_debugfs_open(struct inode *inode, struct file *file) +{ + struct ips_debugfs_node *node = inode->i_private; + + return single_open(file, node->show, node->ips); +} + +static const struct file_operations ips_debugfs_ops = { + .owner = THIS_MODULE, + .open = ips_debugfs_open, + .read = seq_read, + .llseek = seq_lseek, + .release = single_release, +}; + +static void ips_debugfs_cleanup(struct ips_driver *ips) +{ + if (ips->debug_root) + debugfs_remove_recursive(ips->debug_root); + return; +} + +static void ips_debugfs_init(struct ips_driver *ips) +{ + int i; + + ips->debug_root = debugfs_create_dir("ips", NULL); + if (!ips->debug_root) { + dev_err(&ips->dev->dev, + "failed to create debugfs entries: %ld\n", + PTR_ERR(ips->debug_root)); + return; + } + + for (i = 0; i < ARRAY_SIZE(ips_debug_files); i++) { + struct dentry *ent; + struct ips_debugfs_node *node = &ips_debug_files[i]; + + node->ips = ips; + ent = debugfs_create_file(node->name, S_IFREG | S_IRUGO, + ips->debug_root, node, + &ips_debugfs_ops); + if (!ent) { + dev_err(&ips->dev->dev, + "failed to create debug file: %ld\n", + PTR_ERR(ent)); + goto err_cleanup; + } + } + + return; + +err_cleanup: + ips_debugfs_cleanup(ips); + return; +} +#endif /* CONFIG_DEBUG_FS */ + +/** + * ips_detect_cpu - detect whether CPU supports IPS + * + * Walk our list and see if we're on a supported CPU. If we find one, + * return the limits for it. + */ +static struct ips_mcp_limits *ips_detect_cpu(struct ips_driver *ips) +{ + u64 turbo_power, misc_en; + struct ips_mcp_limits *limits = NULL; + u16 tdp; + + if (!(boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model == 37)) { + dev_info(&ips->dev->dev, "Non-IPS CPU detected.\n"); + goto out; + } + + rdmsrl(IA32_MISC_ENABLE, misc_en); + /* + * If the turbo enable bit isn't set, we shouldn't try to enable/disable + * turbo manually or we'll get an illegal MSR access, even though + * turbo will still be available. + */ + if (!(misc_en & IA32_MISC_TURBO_EN)) + ; /* add turbo MSR write allowed flag if necessary */ + + if (strstr(boot_cpu_data.x86_model_id, "CPU M")) + limits = &ips_sv_limits; + else if (strstr(boot_cpu_data.x86_model_id, "CPU L")) + limits = &ips_lv_limits; + else if (strstr(boot_cpu_data.x86_model_id, "CPU U")) + limits = &ips_ulv_limits; + else + dev_info(&ips->dev->dev, "No CPUID match found.\n"); + + rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_power); + tdp = turbo_power & TURBO_TDP_MASK; + + /* Sanity check TDP against CPU */ + if (limits->mcp_power_limit != (tdp / 8) * 1000) { + dev_warn(&ips->dev->dev, "Warning: CPU TDP doesn't match expected value (found %d, expected %d)\n", + tdp / 8, limits->mcp_power_limit / 1000); + } + +out: + return limits; +} + +/** + * ips_get_i915_syms - try to get GPU control methods from i915 driver + * @ips: IPS driver + * + * The i915 driver exports several interfaces to allow the IPS driver to + * monitor and control graphics turbo mode. If we can find them, we can + * enable graphics turbo, otherwise we must disable it to avoid exceeding + * thermal and power limits in the MCP. + */ +static bool ips_get_i915_syms(struct ips_driver *ips) +{ + ips->read_mch_val = symbol_get(i915_read_mch_val); + if (!ips->read_mch_val) + goto out_err; + ips->gpu_raise = symbol_get(i915_gpu_raise); + if (!ips->gpu_raise) + goto out_put_mch; + ips->gpu_lower = symbol_get(i915_gpu_lower); + if (!ips->gpu_lower) + goto out_put_raise; + ips->gpu_busy = symbol_get(i915_gpu_busy); + if (!ips->gpu_busy) + goto out_put_lower; + ips->gpu_turbo_disable = symbol_get(i915_gpu_turbo_disable); + if (!ips->gpu_turbo_disable) + goto out_put_busy; + + return true; + +out_put_busy: + symbol_put(i915_gpu_turbo_disable); +out_put_lower: + symbol_put(i915_gpu_lower); +out_put_raise: + symbol_put(i915_gpu_raise); +out_put_mch: + symbol_put(i915_read_mch_val); +out_err: + return false; +} + +static DEFINE_PCI_DEVICE_TABLE(ips_id_table) = { + { PCI_DEVICE(PCI_VENDOR_ID_INTEL, + PCI_DEVICE_ID_INTEL_THERMAL_SENSOR), }, + { 0, } +}; + +MODULE_DEVICE_TABLE(pci, ips_id_table); + +static int ips_probe(struct pci_dev *dev, const struct pci_device_id *id) +{ + u64 platform_info; + struct ips_driver *ips; + u32 hts; + int ret = 0; + u16 htshi, trc, trc_required_mask; + u8 tse; + + ips = kzalloc(sizeof(struct ips_driver), GFP_KERNEL); + if (!ips) + return -ENOMEM; + + pci_set_drvdata(dev, ips); + ips->dev = dev; + + ips->limits = ips_detect_cpu(ips); + if (!ips->limits) { + dev_info(&dev->dev, "IPS not supported on this CPU\n"); + ret = -ENXIO; + goto error_free; + } + + spin_lock_init(&ips->turbo_status_lock); + + if (!pci_resource_start(dev, 0)) { + dev_err(&dev->dev, "TBAR not assigned, aborting\n"); + ret = -ENXIO; + goto error_free; + } + + ret = pci_request_regions(dev, "ips thermal sensor"); + if (ret) { + dev_err(&dev->dev, "thermal resource busy, aborting\n"); + goto error_free; + } + + ret = pci_enable_device(dev); + if (ret) { + dev_err(&dev->dev, "can't enable PCI device, aborting\n"); + goto error_free; + } + + ips->regmap = ioremap(pci_resource_start(dev, 0), + pci_resource_len(dev, 0)); + if (!ips->regmap) { + dev_err(&dev->dev, "failed to map thermal regs, aborting\n"); + ret = -EBUSY; + goto error_release; + } + + tse = thm_readb(THM_TSE); + if (tse != TSE_EN) { + dev_err(&dev->dev, "thermal device not enabled (0x%02x), aborting\n", tse); + ret = -ENXIO; + goto error_unmap; + } + + trc = thm_readw(THM_TRC); + trc_required_mask = TRC_CORE1_EN | TRC_CORE_PWR | TRC_MCH_EN; + if ((trc & trc_required_mask) != trc_required_mask) { + dev_err(&dev->dev, "thermal reporting for required devices not enabled, aborting\n"); + ret = -ENXIO; + goto error_unmap; + } + + if (trc & TRC_CORE2_EN) + ips->second_cpu = true; + + if (!ips_get_i915_syms(ips)) { + dev_err(&dev->dev, "failed to get i915 symbols, graphics turbo disabled\n"); + ips->gpu_turbo_enabled = false; + } else { + dev_dbg(&dev->dev, "graphics turbo enabled\n"); + ips->gpu_turbo_enabled = true; + } + + update_turbo_limits(ips); + dev_dbg(&dev->dev, "max cpu power clamp: %dW\n", + ips->mcp_power_limit / 10); + dev_dbg(&dev->dev, "max core power clamp: %dW\n", + ips->core_power_limit / 10); + /* BIOS may update limits at runtime */ + if (thm_readl(THM_PSC) & PSP_PBRT) + ips->poll_turbo_status = true; + + /* + * Check PLATFORM_INFO MSR to make sure this chip is + * turbo capable. + */ + rdmsrl(PLATFORM_INFO, platform_info); + if (!(platform_info & PLATFORM_TDP)) { + dev_err(&dev->dev, "platform indicates TDP override unavailable, aborting\n"); + ret = -ENODEV; + goto error_unmap; + } + + /* + * IRQ handler for ME interaction + * Note: don't use MSI here as the PCH has bugs. + */ + pci_disable_msi(dev); + ret = request_irq(dev->irq, ips_irq_handler, IRQF_SHARED, "ips", + ips); + if (ret) { + dev_err(&dev->dev, "request irq failed, aborting\n"); + goto error_unmap; + } + + /* Enable aux, hot & critical interrupts */ + thm_writeb(THM_TSPIEN, TSPIEN_AUX2_LOHI | TSPIEN_CRIT_LOHI | + TSPIEN_HOT_LOHI | TSPIEN_AUX_LOHI); + thm_writeb(THM_TEN, TEN_UPDATE_EN); + + /* Collect adjustment values */ + ips->cta_val = thm_readw(THM_CTA); + ips->pta_val = thm_readw(THM_PTA); + ips->mgta_val = thm_readw(THM_MGTA); + + /* Save turbo limits & ratios */ + rdmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit); + + ips_enable_cpu_turbo(ips); + ips->cpu_turbo_enabled = true; + + /* Set up the work queue and monitor/adjust threads */ + ips->monitor = kthread_run(ips_monitor, ips, "ips-monitor"); + if (IS_ERR(ips->monitor)) { + dev_err(&dev->dev, + "failed to create thermal monitor thread, aborting\n"); + ret = -ENOMEM; + goto error_free_irq; + } + + ips->adjust = kthread_create(ips_adjust, ips, "ips-adjust"); + if (IS_ERR(ips->adjust)) { + dev_err(&dev->dev, + "failed to create thermal adjust thread, aborting\n"); + ret = -ENOMEM; + goto error_thread_cleanup; + } + + hts = (ips->core_power_limit << HTS_PCPL_SHIFT) | + (ips->mcp_temp_limit << HTS_PTL_SHIFT) | HTS_NVV; + htshi = HTS2_PRST_RUNNING << HTS2_PRST_SHIFT; + + thm_writew(THM_HTSHI, htshi); + thm_writel(THM_HTS, hts); + + ips_debugfs_init(ips); + + dev_info(&dev->dev, "IPS driver initialized, MCP temp limit %d\n", + ips->mcp_temp_limit); + return ret; + +error_thread_cleanup: + kthread_stop(ips->monitor); +error_free_irq: + free_irq(ips->dev->irq, ips); +error_unmap: + iounmap(ips->regmap); +error_release: + pci_release_regions(dev); +error_free: + kfree(ips); + return ret; +} + +static void ips_remove(struct pci_dev *dev) +{ + struct ips_driver *ips = pci_get_drvdata(dev); + u64 turbo_override; + + if (!ips) + return; + + ips_debugfs_cleanup(ips); + + /* Release i915 driver */ + if (ips->read_mch_val) + symbol_put(i915_read_mch_val); + if (ips->gpu_raise) + symbol_put(i915_gpu_raise); + if (ips->gpu_lower) + symbol_put(i915_gpu_lower); + if (ips->gpu_busy) + symbol_put(i915_gpu_busy); + if (ips->gpu_turbo_disable) + symbol_put(i915_gpu_turbo_disable); + + rdmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override); + turbo_override &= ~(TURBO_TDC_OVR_EN | TURBO_TDP_OVR_EN); + wrmsrl(TURBO_POWER_CURRENT_LIMIT, turbo_override); + wrmsrl(TURBO_POWER_CURRENT_LIMIT, ips->orig_turbo_limit); + + free_irq(ips->dev->irq, ips); + if (ips->adjust) + kthread_stop(ips->adjust); + if (ips->monitor) + kthread_stop(ips->monitor); + iounmap(ips->regmap); + pci_release_regions(dev); + kfree(ips); + dev_dbg(&dev->dev, "IPS driver removed\n"); +} + +#ifdef CONFIG_PM +static int ips_suspend(struct pci_dev *dev, pm_message_t state) +{ + return 0; +} + +static int ips_resume(struct pci_dev *dev) +{ + return 0; +} +#else +#define ips_suspend NULL +#define ips_resume NULL +#endif /* CONFIG_PM */ + +static void ips_shutdown(struct pci_dev *dev) +{ +} + +static struct pci_driver ips_pci_driver = { + .name = "intel ips", + .id_table = ips_id_table, + .probe = ips_probe, + .remove = ips_remove, + .suspend = ips_suspend, + .resume = ips_resume, + .shutdown = ips_shutdown, +}; + +static int __init ips_init(void) +{ + return pci_register_driver(&ips_pci_driver); +} +module_init(ips_init); + +static void ips_exit(void) +{ + pci_unregister_driver(&ips_pci_driver); + return; +} +module_exit(ips_exit); + +MODULE_LICENSE("GPL"); +MODULE_AUTHOR("Jesse Barnes "); +MODULE_DESCRIPTION("Intelligent Power Sharing Driver"); diff --git a/include/drm/i915_drm.h b/include/drm/i915_drm.h index 7f0028e1010b..8f8b072c4c7b 100644 --- a/include/drm/i915_drm.h +++ b/include/drm/i915_drm.h @@ -33,6 +33,15 @@ * subject to backwards-compatibility constraints. */ +#ifdef __KERNEL__ +/* For use by IPS driver */ +extern unsigned long i915_read_mch_val(void); +extern bool i915_gpu_raise(void); +extern bool i915_gpu_lower(void); +extern bool i915_gpu_busy(void); +extern bool i915_gpu_turbo_disable(void); +#endif + /* Each region is a minimum of 16k, and there are at most 255 of them. */ #define I915_NR_TEX_REGIONS 255 /* table size 2k - maximum due to use -- cgit v1.2.3-70-g09d2 From c715a38bb7fc22fb8018b916c8a9f7ff017a8ad7 Mon Sep 17 00:00:00 2001 From: Alan Cox Date: Fri, 18 Jun 2010 14:05:52 +0100 Subject: rar: Move the RAR driver into the right place as its now clean We exit staging rar! rar! rar!... Signed-off-by: Alan Cox Signed-off-by: Matthew Garrett Acked-by: Greg Kroah-Hartman --- drivers/platform/x86/Kconfig | 22 + drivers/platform/x86/Makefile | 1 + drivers/platform/x86/intel_rar_register.c | 671 +++++++++++++++++++++++++++ drivers/staging/Kconfig | 2 - drivers/staging/Makefile | 1 - drivers/staging/memrar/memrar_handler.c | 3 +- drivers/staging/rar_register/Kconfig | 30 -- drivers/staging/rar_register/Makefile | 2 - drivers/staging/rar_register/rar_register.c | 675 ---------------------------- drivers/staging/rar_register/rar_register.h | 44 -- include/linux/rar_register.h | 44 ++ 11 files changed, 739 insertions(+), 756 deletions(-) create mode 100644 drivers/platform/x86/intel_rar_register.c delete mode 100644 drivers/staging/rar_register/Kconfig delete mode 100644 drivers/staging/rar_register/Makefile delete mode 100644 drivers/staging/rar_register/rar_register.c delete mode 100644 drivers/staging/rar_register/rar_register.h create mode 100644 include/linux/rar_register.h (limited to 'include') diff --git a/drivers/platform/x86/Kconfig b/drivers/platform/x86/Kconfig index 724b2ed1a3cb..2f173bc0ff05 100644 --- a/drivers/platform/x86/Kconfig +++ b/drivers/platform/x86/Kconfig @@ -539,6 +539,28 @@ config INTEL_SCU_IPC some embedded Intel x86 platforms. This is not needed for PC-type machines. +config RAR_REGISTER + bool "Restricted Access Region Register Driver" + depends on PCI && X86_MRST + default n + ---help--- + This driver allows other kernel drivers access to the + contents of the restricted access region control registers. + + The restricted access region control registers + (rar_registers) are used to pass address and + locking information on restricted access regions + to other drivers that use restricted access regions. + + The restricted access regions are regions of memory + on the Intel MID Platform that are not accessible to + the x86 processor, but are accessible to dedicated + processors on board peripheral devices. + + The purpose of the restricted access regions is to + protect sensitive data from compromise by unauthorized + programs running on the x86 processor. + config INTEL_IPS tristate "Intel Intelligent Power Sharing" depends on ACPI diff --git a/drivers/platform/x86/Makefile b/drivers/platform/x86/Makefile index 7318fc2c1629..ed50eca1b556 100644 --- a/drivers/platform/x86/Makefile +++ b/drivers/platform/x86/Makefile @@ -26,4 +26,5 @@ obj-$(CONFIG_TOPSTAR_LAPTOP) += topstar-laptop.o obj-$(CONFIG_ACPI_TOSHIBA) += toshiba_acpi.o obj-$(CONFIG_TOSHIBA_BT_RFKILL) += toshiba_bluetooth.o obj-$(CONFIG_INTEL_SCU_IPC) += intel_scu_ipc.o +obj-$(CONFIG_RAR_REGISTER) += intel_rar_register.o obj-$(CONFIG_INTEL_IPS) += intel_ips.o diff --git a/drivers/platform/x86/intel_rar_register.c b/drivers/platform/x86/intel_rar_register.c new file mode 100644 index 000000000000..73f8e6d72669 --- /dev/null +++ b/drivers/platform/x86/intel_rar_register.c @@ -0,0 +1,671 @@ +/* + * rar_register.c - An Intel Restricted Access Region register driver + * + * Copyright(c) 2009 Intel Corporation. All rights reserved. + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License as + * published by the Free Software Foundation; either version 2 of the + * License, or (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA + * 02111-1307, USA. + * + * ------------------------------------------------------------------- + * 20091204 Mark Allyn + * Ossama Othman + * Cleanup per feedback from Alan Cox and Arjan Van De Ven + * + * 20090806 Ossama Othman + * Return zero high address if upper 22 bits is zero. + * Cleaned up checkpatch errors. + * Clarified that driver is dealing with bus addresses. + * + * 20090702 Ossama Othman + * Removed unnecessary include directives + * Cleaned up spinlocks. + * Cleaned up logging. + * Improved invalid parameter checks. + * Fixed and simplified RAR address retrieval and RAR locking + * code. + * + * 20090626 Mark Allyn + * Initial publish + */ + +#include +#include +#include +#include +#include +#include + +/* === Lincroft Message Bus Interface === */ +#define LNC_MCR_OFFSET 0xD0 /* Message Control Register */ +#define LNC_MDR_OFFSET 0xD4 /* Message Data Register */ + +/* Message Opcodes */ +#define LNC_MESSAGE_READ_OPCODE 0xD0 +#define LNC_MESSAGE_WRITE_OPCODE 0xE0 + +/* Message Write Byte Enables */ +#define LNC_MESSAGE_BYTE_WRITE_ENABLES 0xF + +/* B-unit Port */ +#define LNC_BUNIT_PORT 0x3 + +/* === Lincroft B-Unit Registers - Programmed by IA32 firmware === */ +#define LNC_BRAR0L 0x10 +#define LNC_BRAR0H 0x11 +#define LNC_BRAR1L 0x12 +#define LNC_BRAR1H 0x13 +/* Reserved for SeP */ +#define LNC_BRAR2L 0x14 +#define LNC_BRAR2H 0x15 + +/* Moorestown supports three restricted access regions. */ +#define MRST_NUM_RAR 3 + +/* RAR Bus Address Range */ +struct rar_addr { + dma_addr_t low; + dma_addr_t high; +}; + +/* + * We create one of these for each RAR + */ +struct client { + int (*callback)(unsigned long data); + unsigned long driver_priv; + bool busy; +}; + +static DEFINE_MUTEX(rar_mutex); +static DEFINE_MUTEX(lnc_reg_mutex); + +/* + * One per RAR device (currently only one device) + */ +struct rar_device { + struct rar_addr rar_addr[MRST_NUM_RAR]; + struct pci_dev *rar_dev; + bool registered; + bool allocated; + struct client client[MRST_NUM_RAR]; +}; + +/* Current platforms have only one rar_device for 3 rar regions */ +static struct rar_device my_rar_device; + +/* + * Abstract out multiple device support. Current platforms only + * have a single RAR device. + */ + +/** + * alloc_rar_device - return a new RAR structure + * + * Return a new (but not yet ready) RAR device object + */ +static struct rar_device *alloc_rar_device(void) +{ + if (my_rar_device.allocated) + return NULL; + my_rar_device.allocated = 1; + return &my_rar_device; +} + +/** + * free_rar_device - free a RAR object + * @rar: the RAR device being freed + * + * Release a RAR object and any attached resources + */ +static void free_rar_device(struct rar_device *rar) +{ + pci_dev_put(rar->rar_dev); + rar->allocated = 0; +} + +/** + * _rar_to_device - return the device handling this RAR + * @rar: RAR number + * @off: returned offset + * + * Internal helper for looking up RAR devices. This and alloc are the + * two functions that need touching to go to multiple RAR devices. + */ +static struct rar_device *_rar_to_device(int rar, int *off) +{ + if (rar >= 0 && rar <= 3) { + *off = rar; + return &my_rar_device; + } + return NULL; +} + +/** + * rar_to_device - return the device handling this RAR + * @rar: RAR number + * @off: returned offset + * + * Return the device this RAR maps to if one is present, otherwise + * returns NULL. Reports the offset relative to the base of this + * RAR device in off. + */ +static struct rar_device *rar_to_device(int rar, int *off) +{ + struct rar_device *rar_dev = _rar_to_device(rar, off); + if (rar_dev == NULL || !rar_dev->registered) + return NULL; + return rar_dev; +} + +/** + * rar_to_client - return the client handling this RAR + * @rar: RAR number + * + * Return the client this RAR maps to if a mapping is known, otherwise + * returns NULL. + */ +static struct client *rar_to_client(int rar) +{ + int idx; + struct rar_device *r = _rar_to_device(rar, &idx); + if (r != NULL) + return &r->client[idx]; + return NULL; +} + +/** + * rar_read_addr - retrieve a RAR mapping + * @pdev: PCI device for the RAR + * @offset: offset for message + * @addr: returned address + * + * Reads the address of a given RAR register. Returns 0 on success + * or an error code on failure. + */ +static int rar_read_addr(struct pci_dev *pdev, int offset, dma_addr_t *addr) +{ + /* + * ======== The Lincroft Message Bus Interface ======== + * Lincroft registers may be obtained via PCI from + * the host bridge using the Lincroft Message Bus + * Interface. That message bus interface is generally + * comprised of two registers: a control register (MCR, 0xDO) + * and a data register (MDR, 0xD4). + * + * The MCR (message control register) format is the following: + * 1. [31:24]: Opcode + * 2. [23:16]: Port + * 3. [15:8]: Register Offset + * 4. [7:4]: Byte Enables (use 0xF to set all of these bits + * to 1) + * 5. [3:0]: reserved + * + * Read (0xD0) and write (0xE0) opcodes are written to the + * control register when reading and writing to Lincroft + * registers, respectively. + * + * We're interested in registers found in the Lincroft + * B-unit. The B-unit port is 0x3. + * + * The six B-unit RAR register offsets we use are listed + * earlier in this file. + * + * Lastly writing to the MCR register requires the "Byte + * enables" bits to be set to 1. This may be achieved by + * writing 0xF at bit 4. + * + * The MDR (message data register) format is the following: + * 1. [31:0]: Read/Write Data + * + * Data being read from this register is only available after + * writing the appropriate control message to the MCR + * register. + * + * Data being written to this register must be written before + * writing the appropriate control message to the MCR + * register. + */ + + int result; + u32 addr32; + + /* Construct control message */ + u32 const message = + (LNC_MESSAGE_READ_OPCODE << 24) + | (LNC_BUNIT_PORT << 16) + | (offset << 8) + | (LNC_MESSAGE_BYTE_WRITE_ENABLES << 4); + + dev_dbg(&pdev->dev, "Offset for 'get' LNC MSG is %x\n", offset); + + /* + * We synchronize access to the Lincroft MCR and MDR registers + * until BOTH the command is issued through the MCR register + * and the corresponding data is read from the MDR register. + * Otherwise a race condition would exist between accesses to + * both registers. + */ + + mutex_lock(&lnc_reg_mutex); + + /* Send the control message */ + result = pci_write_config_dword(pdev, LNC_MCR_OFFSET, message); + if (!result) { + /* Read back the address as a 32bit value */ + result = pci_read_config_dword(pdev, LNC_MDR_OFFSET, &addr32); + *addr = (dma_addr_t)addr32; + } + mutex_unlock(&lnc_reg_mutex); + return result; +} + +/** + * rar_set_addr - Set a RAR mapping + * @pdev: PCI device for the RAR + * @offset: offset for message + * @addr: address to set + * + * Sets the address of a given RAR register. Returns 0 on success + * or an error code on failure. + */ +static int rar_set_addr(struct pci_dev *pdev, + int offset, + dma_addr_t addr) +{ + /* + * Data being written to this register must be written before + * writing the appropriate control message to the MCR + * register. + * See rar_get_addrs() for a description of the + * message bus interface being used here. + */ + + int result; + + /* Construct control message */ + u32 const message = (LNC_MESSAGE_WRITE_OPCODE << 24) + | (LNC_BUNIT_PORT << 16) + | (offset << 8) + | (LNC_MESSAGE_BYTE_WRITE_ENABLES << 4); + + /* + * We synchronize access to the Lincroft MCR and MDR registers + * until BOTH the command is issued through the MCR register + * and the corresponding data is read from the MDR register. + * Otherwise a race condition would exist between accesses to + * both registers. + */ + + mutex_lock(&lnc_reg_mutex); + + /* Send the control message */ + result = pci_write_config_dword(pdev, LNC_MDR_OFFSET, addr); + if (!result) + /* And address */ + result = pci_write_config_dword(pdev, LNC_MCR_OFFSET, message); + + mutex_unlock(&lnc_reg_mutex); + return result; +} + +/* + * rar_init_params - Initialize RAR parameters + * @rar: RAR device to initialise + * + * Initialize RAR parameters, such as bus addresses, etc. Returns 0 + * on success, or an error code on failure. + */ +static int init_rar_params(struct rar_device *rar) +{ + struct pci_dev *pdev = rar->rar_dev; + unsigned int i; + int result = 0; + int offset = 0x10; /* RAR 0 to 2 in order low/high/low/high/... */ + + /* Retrieve RAR start and end bus addresses. + * Access the RAR registers through the Lincroft Message Bus + * Interface on PCI device: 00:00.0 Host bridge. + */ + + for (i = 0; i < MRST_NUM_RAR; ++i) { + struct rar_addr *addr = &rar->rar_addr[i]; + + result = rar_read_addr(pdev, offset++, &addr->low); + if (result != 0) + return result; + + result = rar_read_addr(pdev, offset++, &addr->high); + if (result != 0) + return result; + + + /* + * Only the upper 22 bits of the RAR addresses are + * stored in their corresponding RAR registers so we + * must set the lower 10 bits accordingly. + + * The low address has its lower 10 bits cleared, and + * the high address has all its lower 10 bits set, + * e.g.: + * low = 0x2ffffc00 + */ + + addr->low &= (dma_addr_t)0xfffffc00u; + + /* + * Set bits 9:0 on uppser address if bits 31:10 are non + * zero; otherwize clear all bits + */ + + if ((addr->high & 0xfffffc00u) == 0) + addr->high = 0; + else + addr->high |= 0x3ffu; + } + /* Done accessing the device. */ + + if (result == 0) { + for (i = 0; i != MRST_NUM_RAR; ++i) { + /* + * "BRAR" refers to the RAR registers in the + * Lincroft B-unit. + */ + dev_info(&pdev->dev, "BRAR[%u] bus address range = " + "[%lx, %lx]\n", i, + (unsigned long)rar->rar_addr[i].low, + (unsigned long)rar->rar_addr[i].high); + } + } + return result; +} + +/** + * rar_get_address - get the bus address in a RAR + * @start: return value of start address of block + * @end: return value of end address of block + * + * The rar_get_address function is used by other device drivers + * to obtain RAR address information on a RAR. It takes three + * parameters: + * + * The function returns a 0 upon success or an error if there is no RAR + * facility on this system. + */ +int rar_get_address(int rar_index, dma_addr_t *start, dma_addr_t *end) +{ + int idx; + struct rar_device *rar = rar_to_device(rar_index, &idx); + + if (rar == NULL) { + WARN_ON(1); + return -ENODEV; + } + + *start = rar->rar_addr[idx].low; + *end = rar->rar_addr[idx].high; + return 0; +} +EXPORT_SYMBOL(rar_get_address); + +/** + * rar_lock - lock a RAR register + * @rar_index: RAR to lock (0-2) + * + * The rar_lock function is ued by other device drivers to lock an RAR. + * once a RAR is locked, it stays locked until the next system reboot. + * + * The function returns a 0 upon success or an error if there is no RAR + * facility on this system, or the locking fails + */ +int rar_lock(int rar_index) +{ + struct rar_device *rar; + int result; + int idx; + dma_addr_t low, high; + + rar = rar_to_device(rar_index, &idx); + + if (rar == NULL) { + WARN_ON(1); + return -EINVAL; + } + + low = rar->rar_addr[idx].low & 0xfffffc00u; + high = rar->rar_addr[idx].high & 0xfffffc00u; + + /* + * Only allow I/O from the graphics and Langwell; + * not from the x86 processor + */ + + if (rar_index == RAR_TYPE_VIDEO) { + low |= 0x00000009; + high |= 0x00000015; + } else if (rar_index == RAR_TYPE_AUDIO) { + /* Only allow I/O from Langwell; nothing from x86 */ + low |= 0x00000008; + high |= 0x00000018; + } else + /* Read-only from all agents */ + high |= 0x00000018; + + /* + * Now program the register using the Lincroft message + * bus interface. + */ + result = rar_set_addr(rar->rar_dev, + 2 * idx, low); + + if (result == 0) + result = rar_set_addr(rar->rar_dev, + 2 * idx + 1, high); + + return result; +} +EXPORT_SYMBOL(rar_lock); + +/** + * register_rar - register a RAR handler + * @num: RAR we wish to register for + * @callback: function to call when RAR support is available + * @data: data to pass to this function + * + * The register_rar function is to used by other device drivers + * to ensure that this driver is ready. As we cannot be sure of + * the compile/execute order of drivers in ther kernel, it is + * best to give this driver a callback function to call when + * it is ready to give out addresses. The callback function + * would have those steps that continue the initialization of + * a driver that do require a valid RAR address. One of those + * steps would be to call rar_get_address() + * + * This function return 0 on success or an error code on failure. + */ +int register_rar(int num, int (*callback)(unsigned long data), + unsigned long data) +{ + /* For now we hardcode a single RAR device */ + struct rar_device *rar; + struct client *c; + int idx; + int retval = 0; + + mutex_lock(&rar_mutex); + + /* Do we have a client mapping for this RAR number ? */ + c = rar_to_client(num); + if (c == NULL) { + retval = -ERANGE; + goto done; + } + /* Is it claimed ? */ + if (c->busy) { + retval = -EBUSY; + goto done; + } + c->busy = 1; + + /* See if we have a handler for this RAR yet, if we do then fire it */ + rar = rar_to_device(num, &idx); + + if (rar) { + /* + * if the driver already registered, then we can simply + * call the callback right now + */ + (*callback)(data); + goto done; + } + + /* Arrange to be called back when the hardware is found */ + c->callback = callback; + c->driver_priv = data; +done: + mutex_unlock(&rar_mutex); + return retval; +} +EXPORT_SYMBOL(register_rar); + +/** + * unregister_rar - release a RAR allocation + * @num: RAR number + * + * Releases a RAR allocation, or pending allocation. If a callback is + * pending then this function will either complete before the unregister + * returns or not at all. + */ + +void unregister_rar(int num) +{ + struct client *c; + + mutex_lock(&rar_mutex); + c = rar_to_client(num); + if (c == NULL || !c->busy) + WARN_ON(1); + else + c->busy = 0; + mutex_unlock(&rar_mutex); +} +EXPORT_SYMBOL(unregister_rar); + +/** + * rar_callback - Process callbacks + * @rar: new RAR device + * + * Process the callbacks for a newly found RAR device. + */ + +static void rar_callback(struct rar_device *rar) +{ + struct client *c = &rar->client[0]; + int i; + + mutex_lock(&rar_mutex); + + rar->registered = 1; /* Ensure no more callbacks queue */ + + for (i = 0; i < MRST_NUM_RAR; i++) { + if (c->callback && c->busy) { + c->callback(c->driver_priv); + c->callback = NULL; + } + c++; + } + mutex_unlock(&rar_mutex); +} + +/** + * rar_probe - PCI probe callback + * @dev: PCI device + * @id: matching entry in the match table + * + * A RAR device has been discovered. Initialise it and if successful + * process any pending callbacks that can now be completed. + */ +static int rar_probe(struct pci_dev *dev, const struct pci_device_id *id) +{ + int error; + struct rar_device *rar; + + dev_dbg(&dev->dev, "PCI probe starting\n"); + + rar = alloc_rar_device(); + if (rar == NULL) + return -EBUSY; + + /* Enable the device */ + error = pci_enable_device(dev); + if (error) { + dev_err(&dev->dev, + "Error enabling RAR register PCI device\n"); + goto end_function; + } + + /* Fill in the rar_device structure */ + rar->rar_dev = pci_dev_get(dev); + pci_set_drvdata(dev, rar); + + /* + * Initialize the RAR parameters, which have to be retrieved + * via the message bus interface. + */ + error = init_rar_params(rar); + if (error) { + pci_disable_device(dev); + dev_err(&dev->dev, "Error retrieving RAR addresses\n"); + goto end_function; + } + /* now call anyone who has registered (using callbacks) */ + rar_callback(rar); + return 0; +end_function: + free_rar_device(rar); + return error; +} + +const struct pci_device_id rar_pci_id_tbl[] = { + { PCI_VDEVICE(INTEL, 0x4110) }, + { 0 } +}; + +MODULE_DEVICE_TABLE(pci, rar_pci_id_tbl); + +const struct pci_device_id *my_id_table = rar_pci_id_tbl; + +/* field for registering driver to PCI device */ +static struct pci_driver rar_pci_driver = { + .name = "rar_register_driver", + .id_table = rar_pci_id_tbl, + .probe = rar_probe, + /* Cannot be unplugged - no remove */ +}; + +static int __init rar_init_handler(void) +{ + return pci_register_driver(&rar_pci_driver); +} + +static void __exit rar_exit_handler(void) +{ + pci_unregister_driver(&rar_pci_driver); +} + +module_init(rar_init_handler); +module_exit(rar_exit_handler); + +MODULE_LICENSE("GPL"); +MODULE_DESCRIPTION("Intel Restricted Access Region Register Driver"); diff --git a/drivers/staging/Kconfig b/drivers/staging/Kconfig index 984a75440710..9dfef8a59974 100644 --- a/drivers/staging/Kconfig +++ b/drivers/staging/Kconfig @@ -109,8 +109,6 @@ source "drivers/staging/hv/Kconfig" source "drivers/staging/vme/Kconfig" -source "drivers/staging/rar_register/Kconfig" - source "drivers/staging/memrar/Kconfig" source "drivers/staging/sep/Kconfig" diff --git a/drivers/staging/Makefile b/drivers/staging/Makefile index 9fa25133874a..3dbf681ca64e 100644 --- a/drivers/staging/Makefile +++ b/drivers/staging/Makefile @@ -35,7 +35,6 @@ obj-$(CONFIG_VT6656) += vt6656/ obj-$(CONFIG_FB_UDL) += udlfb/ obj-$(CONFIG_HYPERV) += hv/ obj-$(CONFIG_VME_BUS) += vme/ -obj-$(CONFIG_RAR_REGISTER) += rar_register/ obj-$(CONFIG_MRST_RAR_HANDLER) += memrar/ obj-$(CONFIG_DX_SEP) += sep/ obj-$(CONFIG_IIO) += iio/ diff --git a/drivers/staging/memrar/memrar_handler.c b/drivers/staging/memrar/memrar_handler.c index efa7fd62d390..41876f2b0e54 100644 --- a/drivers/staging/memrar/memrar_handler.c +++ b/drivers/staging/memrar/memrar_handler.c @@ -47,8 +47,7 @@ #include #include #include - -#include "../rar_register/rar_register.h" +#include #include "memrar.h" #include "memrar_allocator.h" diff --git a/drivers/staging/rar_register/Kconfig b/drivers/staging/rar_register/Kconfig deleted file mode 100644 index e9c27738199b..000000000000 --- a/drivers/staging/rar_register/Kconfig +++ /dev/null @@ -1,30 +0,0 @@ -# -# RAR device configuration -# - -menu "RAR Register Driver" -# -# Restricted Access Register Manager -# -config RAR_REGISTER - tristate "Restricted Access Region Register Driver" - depends on PCI - default n - ---help--- - This driver allows other kernel drivers access to the - contents of the restricted access region control registers. - - The restricted access region control registers - (rar_registers) are used to pass address and - locking information on restricted access regions - to other drivers that use restricted access regions. - - The restricted access regions are regions of memory - on the Intel MID Platform that are not accessible to - the x86 processor, but are accessible to dedicated - processors on board peripheral devices. - - The purpose of the restricted access regions is to - protect sensitive data from compromise by unauthorized - programs running on the x86 processor. -endmenu diff --git a/drivers/staging/rar_register/Makefile b/drivers/staging/rar_register/Makefile deleted file mode 100644 index d5954ccc16c9..000000000000 --- a/drivers/staging/rar_register/Makefile +++ /dev/null @@ -1,2 +0,0 @@ -EXTRA_CFLAGS += -DLITTLE__ENDIAN -obj-$(CONFIG_RAR_REGISTER) += rar_register.o diff --git a/drivers/staging/rar_register/rar_register.c b/drivers/staging/rar_register/rar_register.c deleted file mode 100644 index 618503f422ef..000000000000 --- a/drivers/staging/rar_register/rar_register.c +++ /dev/null @@ -1,675 +0,0 @@ -/* - * rar_register.c - An Intel Restricted Access Region register driver - * - * Copyright(c) 2009 Intel Corporation. All rights reserved. - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public License as - * published by the Free Software Foundation; either version 2 of the - * License, or (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA - * 02111-1307, USA. - * - * ------------------------------------------------------------------- - * 20091204 Mark Allyn - * Ossama Othman - * Cleanup per feedback from Alan Cox and Arjan Van De Ven - * - * 20090806 Ossama Othman - * Return zero high address if upper 22 bits is zero. - * Cleaned up checkpatch errors. - * Clarified that driver is dealing with bus addresses. - * - * 20090702 Ossama Othman - * Removed unnecessary include directives - * Cleaned up spinlocks. - * Cleaned up logging. - * Improved invalid parameter checks. - * Fixed and simplified RAR address retrieval and RAR locking - * code. - * - * 20090626 Mark Allyn - * Initial publish - */ - -#define DEBUG 1 - -#include "rar_register.h" - -#include -#include -#include -#include -#include - -/* === Lincroft Message Bus Interface === */ -#define LNC_MCR_OFFSET 0xD0 /* Message Control Register */ -#define LNC_MDR_OFFSET 0xD4 /* Message Data Register */ - -/* Message Opcodes */ -#define LNC_MESSAGE_READ_OPCODE 0xD0 -#define LNC_MESSAGE_WRITE_OPCODE 0xE0 - -/* Message Write Byte Enables */ -#define LNC_MESSAGE_BYTE_WRITE_ENABLES 0xF - -/* B-unit Port */ -#define LNC_BUNIT_PORT 0x3 - -/* === Lincroft B-Unit Registers - Programmed by IA32 firmware === */ -#define LNC_BRAR0L 0x10 -#define LNC_BRAR0H 0x11 -#define LNC_BRAR1L 0x12 -#define LNC_BRAR1H 0x13 -/* Reserved for SeP */ -#define LNC_BRAR2L 0x14 -#define LNC_BRAR2H 0x15 - -/* Moorestown supports three restricted access regions. */ -#define MRST_NUM_RAR 3 - -/* RAR Bus Address Range */ -struct rar_addr { - dma_addr_t low; - dma_addr_t high; -}; - -/* - * We create one of these for each RAR - */ -struct client { - int (*callback)(unsigned long data); - unsigned long driver_priv; - bool busy; -}; - -static DEFINE_MUTEX(rar_mutex); -static DEFINE_MUTEX(lnc_reg_mutex); - -/* - * One per RAR device (currently only one device) - */ -struct rar_device { - struct rar_addr rar_addr[MRST_NUM_RAR]; - struct pci_dev *rar_dev; - bool registered; - bool allocated; - struct client client[MRST_NUM_RAR]; -}; - -/* Current platforms have only one rar_device for 3 rar regions */ -static struct rar_device my_rar_device; - -/* - * Abstract out multiple device support. Current platforms only - * have a single RAR device. - */ - -/** - * alloc_rar_device - return a new RAR structure - * - * Return a new (but not yet ready) RAR device object - */ -static struct rar_device *alloc_rar_device(void) -{ - if (my_rar_device.allocated) - return NULL; - my_rar_device.allocated = 1; - return &my_rar_device; -} - -/** - * free_rar_device - free a RAR object - * @rar: the RAR device being freed - * - * Release a RAR object and any attached resources - */ -static void free_rar_device(struct rar_device *rar) -{ - pci_dev_put(rar->rar_dev); - rar->allocated = 0; -} - -/** - * _rar_to_device - return the device handling this RAR - * @rar: RAR number - * @off: returned offset - * - * Internal helper for looking up RAR devices. This and alloc are the - * two functions that need touching to go to multiple RAR devices. - */ -static struct rar_device *_rar_to_device(int rar, int *off) -{ - if (rar >= 0 && rar <= 3) { - *off = rar; - return &my_rar_device; - } - return NULL; -} - - -/** - * rar_to_device - return the device handling this RAR - * @rar: RAR number - * @off: returned offset - * - * Return the device this RAR maps to if one is present, otherwise - * returns NULL. Reports the offset relative to the base of this - * RAR device in off. - */ -static struct rar_device *rar_to_device(int rar, int *off) -{ - struct rar_device *rar_dev = _rar_to_device(rar, off); - if (rar_dev == NULL || !rar_dev->registered) - return NULL; - return rar_dev; -} - -/** - * rar_to_client - return the client handling this RAR - * @rar: RAR number - * - * Return the client this RAR maps to if a mapping is known, otherwise - * returns NULL. - */ -static struct client *rar_to_client(int rar) -{ - int idx; - struct rar_device *r = _rar_to_device(rar, &idx); - if (r != NULL) - return &r->client[idx]; - return NULL; -} - -/** - * rar_read_addr - retrieve a RAR mapping - * @pdev: PCI device for the RAR - * @offset: offset for message - * @addr: returned address - * - * Reads the address of a given RAR register. Returns 0 on success - * or an error code on failure. - */ -static int rar_read_addr(struct pci_dev *pdev, int offset, dma_addr_t *addr) -{ - /* - * ======== The Lincroft Message Bus Interface ======== - * Lincroft registers may be obtained via PCI from - * the host bridge using the Lincroft Message Bus - * Interface. That message bus interface is generally - * comprised of two registers: a control register (MCR, 0xDO) - * and a data register (MDR, 0xD4). - * - * The MCR (message control register) format is the following: - * 1. [31:24]: Opcode - * 2. [23:16]: Port - * 3. [15:8]: Register Offset - * 4. [7:4]: Byte Enables (use 0xF to set all of these bits - * to 1) - * 5. [3:0]: reserved - * - * Read (0xD0) and write (0xE0) opcodes are written to the - * control register when reading and writing to Lincroft - * registers, respectively. - * - * We're interested in registers found in the Lincroft - * B-unit. The B-unit port is 0x3. - * - * The six B-unit RAR register offsets we use are listed - * earlier in this file. - * - * Lastly writing to the MCR register requires the "Byte - * enables" bits to be set to 1. This may be achieved by - * writing 0xF at bit 4. - * - * The MDR (message data register) format is the following: - * 1. [31:0]: Read/Write Data - * - * Data being read from this register is only available after - * writing the appropriate control message to the MCR - * register. - * - * Data being written to this register must be written before - * writing the appropriate control message to the MCR - * register. - */ - - int result; - u32 addr32; - - /* Construct control message */ - u32 const message = - (LNC_MESSAGE_READ_OPCODE << 24) - | (LNC_BUNIT_PORT << 16) - | (offset << 8) - | (LNC_MESSAGE_BYTE_WRITE_ENABLES << 4); - - dev_dbg(&pdev->dev, "Offset for 'get' LNC MSG is %x\n", offset); - - /* - * We synchronize access to the Lincroft MCR and MDR registers - * until BOTH the command is issued through the MCR register - * and the corresponding data is read from the MDR register. - * Otherwise a race condition would exist between accesses to - * both registers. - */ - - mutex_lock(&lnc_reg_mutex); - - /* Send the control message */ - result = pci_write_config_dword(pdev, LNC_MCR_OFFSET, message); - if (!result) { - /* Read back the address as a 32bit value */ - result = pci_read_config_dword(pdev, LNC_MDR_OFFSET, &addr32); - *addr = (dma_addr_t)addr32; - } - mutex_unlock(&lnc_reg_mutex); - return result; -} - -/** - * rar_set_addr - Set a RAR mapping - * @pdev: PCI device for the RAR - * @offset: offset for message - * @addr: address to set - * - * Sets the address of a given RAR register. Returns 0 on success - * or an error code on failure. - */ -static int rar_set_addr(struct pci_dev *pdev, - int offset, - dma_addr_t addr) -{ - /* - * Data being written to this register must be written before - * writing the appropriate control message to the MCR - * register. - * See rar_get_addrs() for a description of the - * message bus interface being used here. - */ - - int result; - - /* Construct control message */ - u32 const message = (LNC_MESSAGE_WRITE_OPCODE << 24) - | (LNC_BUNIT_PORT << 16) - | (offset << 8) - | (LNC_MESSAGE_BYTE_WRITE_ENABLES << 4); - - /* - * We synchronize access to the Lincroft MCR and MDR registers - * until BOTH the command is issued through the MCR register - * and the corresponding data is read from the MDR register. - * Otherwise a race condition would exist between accesses to - * both registers. - */ - - mutex_lock(&lnc_reg_mutex); - - /* Send the control message */ - result = pci_write_config_dword(pdev, LNC_MDR_OFFSET, addr); - if (!result) - /* And address */ - result = pci_write_config_dword(pdev, LNC_MCR_OFFSET, message); - - mutex_unlock(&lnc_reg_mutex); - return result; -} - -/* - * rar_init_params - Initialize RAR parameters - * @rar: RAR device to initialise - * - * Initialize RAR parameters, such as bus addresses, etc. Returns 0 - * on success, or an error code on failure. - */ -static int init_rar_params(struct rar_device *rar) -{ - struct pci_dev *pdev = rar->rar_dev; - unsigned int i; - int result = 0; - int offset = 0x10; /* RAR 0 to 2 in order low/high/low/high/... */ - - /* Retrieve RAR start and end bus addresses. - * Access the RAR registers through the Lincroft Message Bus - * Interface on PCI device: 00:00.0 Host bridge. - */ - - for (i = 0; i < MRST_NUM_RAR; ++i) { - struct rar_addr *addr = &rar->rar_addr[i]; - - result = rar_read_addr(pdev, offset++, &addr->low); - if (result != 0) - return result; - - result = rar_read_addr(pdev, offset++, &addr->high); - if (result != 0) - return result; - - - /* - * Only the upper 22 bits of the RAR addresses are - * stored in their corresponding RAR registers so we - * must set the lower 10 bits accordingly. - - * The low address has its lower 10 bits cleared, and - * the high address has all its lower 10 bits set, - * e.g.: - * low = 0x2ffffc00 - */ - - addr->low &= (dma_addr_t)0xfffffc00u; - - /* - * Set bits 9:0 on uppser address if bits 31:10 are non - * zero; otherwize clear all bits - */ - - if ((addr->high & 0xfffffc00u) == 0) - addr->high = 0; - else - addr->high |= 0x3ffu; - } - /* Done accessing the device. */ - - if (result == 0) { - for (i = 0; i != MRST_NUM_RAR; ++i) { - /* - * "BRAR" refers to the RAR registers in the - * Lincroft B-unit. - */ - dev_info(&pdev->dev, "BRAR[%u] bus address range = " - "[%lx, %lx]\n", i, - (unsigned long)rar->rar_addr[i].low, - (unsigned long)rar->rar_addr[i].high); - } - } - return result; -} - -/** - * rar_get_address - get the bus address in a RAR - * @start: return value of start address of block - * @end: return value of end address of block - * - * The rar_get_address function is used by other device drivers - * to obtain RAR address information on a RAR. It takes three - * parameters: - * - * The function returns a 0 upon success or an error if there is no RAR - * facility on this system. - */ -int rar_get_address(int rar_index, dma_addr_t *start, dma_addr_t *end) -{ - int idx; - struct rar_device *rar = rar_to_device(rar_index, &idx); - - if (rar == NULL) { - WARN_ON(1); - return -ENODEV; - } - - *start = rar->rar_addr[idx].low; - *end = rar->rar_addr[idx].high; - return 0; -} -EXPORT_SYMBOL(rar_get_address); - -/** - * rar_lock - lock a RAR register - * @rar_index: RAR to lock (0-2) - * - * The rar_lock function is ued by other device drivers to lock an RAR. - * once a RAR is locked, it stays locked until the next system reboot. - * - * The function returns a 0 upon success or an error if there is no RAR - * facility on this system, or the locking fails - */ -int rar_lock(int rar_index) -{ - struct rar_device *rar; - int result; - int idx; - dma_addr_t low, high; - - rar = rar_to_device(rar_index, &idx); - - if (rar == NULL) { - WARN_ON(1); - return -EINVAL; - } - - low = rar->rar_addr[idx].low & 0xfffffc00u; - high = rar->rar_addr[idx].high & 0xfffffc00u; - - /* - * Only allow I/O from the graphics and Langwell; - * not from the x86 processor - */ - - if (rar_index == RAR_TYPE_VIDEO) { - low |= 0x00000009; - high |= 0x00000015; - } else if (rar_index == RAR_TYPE_AUDIO) { - /* Only allow I/O from Langwell; nothing from x86 */ - low |= 0x00000008; - high |= 0x00000018; - } else - /* Read-only from all agents */ - high |= 0x00000018; - - /* - * Now program the register using the Lincroft message - * bus interface. - */ - result = rar_set_addr(rar->rar_dev, - 2 * idx, low); - - if (result == 0) - result = rar_set_addr(rar->rar_dev, - 2 * idx + 1, high); - - return result; -} -EXPORT_SYMBOL(rar_lock); - -/** - * register_rar - register a RAR handler - * @num: RAR we wish to register for - * @callback: function to call when RAR support is available - * @data: data to pass to this function - * - * The register_rar function is to used by other device drivers - * to ensure that this driver is ready. As we cannot be sure of - * the compile/execute order of drivers in ther kernel, it is - * best to give this driver a callback function to call when - * it is ready to give out addresses. The callback function - * would have those steps that continue the initialization of - * a driver that do require a valid RAR address. One of those - * steps would be to call rar_get_address() - * - * This function return 0 on success an error code on failure. - */ -int register_rar(int num, int (*callback)(unsigned long data), - unsigned long data) -{ - /* For now we hardcode a single RAR device */ - struct rar_device *rar; - struct client *c; - int idx; - int retval = 0; - - mutex_lock(&rar_mutex); - - /* Do we have a client mapping for this RAR number ? */ - c = rar_to_client(num); - if (c == NULL) { - retval = -ERANGE; - goto done; - } - /* Is it claimed ? */ - if (c->busy) { - retval = -EBUSY; - goto done; - } - c->busy = 1; - - /* See if we have a handler for this RAR yet, if we do then fire it */ - rar = rar_to_device(num, &idx); - - if (rar) { - /* - * if the driver already registered, then we can simply - * call the callback right now - */ - (*callback)(data); - goto done; - } - - /* Arrange to be called back when the hardware is found */ - c->callback = callback; - c->driver_priv = data; -done: - mutex_unlock(&rar_mutex); - return retval; -} -EXPORT_SYMBOL(register_rar); - -/** - * unregister_rar - release a RAR allocation - * @num: RAR number - * - * Releases a RAR allocation, or pending allocation. If a callback is - * pending then this function will either complete before the unregister - * returns or not at all. - */ - -void unregister_rar(int num) -{ - struct client *c; - - mutex_lock(&rar_mutex); - c = rar_to_client(num); - if (c == NULL || !c->busy) - WARN_ON(1); - else - c->busy = 0; - mutex_unlock(&rar_mutex); -} -EXPORT_SYMBOL(unregister_rar); - -/** - * rar_callback - Process callbacks - * @rar: new RAR device - * - * Process the callbacks for a newly found RAR device. - */ - -static void rar_callback(struct rar_device *rar) -{ - struct client *c = &rar->client[0]; - int i; - - mutex_lock(&rar_mutex); - - rar->registered = 1; /* Ensure no more callbacks queue */ - - for (i = 0; i < MRST_NUM_RAR; i++) { - if (c->callback && c->busy) { - c->callback(c->driver_priv); - c->callback = NULL; - } - c++; - } - mutex_unlock(&rar_mutex); -} - -/** - * rar_probe - PCI probe callback - * @dev: PCI device - * @id: matching entry in the match table - * - * A RAR device has been discovered. Initialise it and if successful - * process any pending callbacks that can now be completed. - */ -static int rar_probe(struct pci_dev *dev, const struct pci_device_id *id) -{ - int error; - struct rar_device *rar; - - dev_dbg(&dev->dev, "PCI probe starting\n"); - - rar = alloc_rar_device(); - if (rar == NULL) - return -EBUSY; - - /* Enable the device */ - error = pci_enable_device(dev); - if (error) { - dev_err(&dev->dev, - "Error enabling RAR register PCI device\n"); - goto end_function; - } - - /* Fill in the rar_device structure */ - rar->rar_dev = pci_dev_get(dev); - pci_set_drvdata(dev, rar); - - /* - * Initialize the RAR parameters, which have to be retrieved - * via the message bus interface. - */ - error = init_rar_params(rar); - if (error) { - pci_disable_device(dev); - dev_err(&dev->dev, "Error retrieving RAR addresses\n"); - goto end_function; - } - /* now call anyone who has registered (using callbacks) */ - rar_callback(rar); - return 0; -end_function: - free_rar_device(rar); - return error; -} - -const struct pci_device_id rar_pci_id_tbl[] = { - { PCI_VDEVICE(INTEL, 0x4110) }, - { 0 } -}; - -MODULE_DEVICE_TABLE(pci, rar_pci_id_tbl); - -const struct pci_device_id *my_id_table = rar_pci_id_tbl; - -/* field for registering driver to PCI device */ -static struct pci_driver rar_pci_driver = { - .name = "rar_register_driver", - .id_table = rar_pci_id_tbl, - .probe = rar_probe, - /* Cannot be unplugged - no remove */ -}; - -static int __init rar_init_handler(void) -{ - return pci_register_driver(&rar_pci_driver); -} - -static void __exit rar_exit_handler(void) -{ - pci_unregister_driver(&rar_pci_driver); -} - -module_init(rar_init_handler); -module_exit(rar_exit_handler); - -MODULE_LICENSE("GPL"); -MODULE_DESCRIPTION("Intel Restricted Access Region Register Driver"); diff --git a/drivers/staging/rar_register/rar_register.h b/drivers/staging/rar_register/rar_register.h deleted file mode 100644 index ffa805780f85..000000000000 --- a/drivers/staging/rar_register/rar_register.h +++ /dev/null @@ -1,44 +0,0 @@ -/* - * Copyright (C) 2010 Intel Corporation. All rights reserved. - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of version 2 of the GNU General - * Public License as published by the Free Software Foundation. - * - * This program is distributed in the hope that it will be - * useful, but WITHOUT ANY WARRANTY; without even the implied - * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR - * PURPOSE. See the GNU General Public License for more details. - * You should have received a copy of the GNU General Public - * License along with this program; if not, write to the Free - * Software Foundation, Inc., 59 Temple Place - Suite 330, - * Boston, MA 02111-1307, USA. - * The full GNU General Public License is included in this - * distribution in the file called COPYING. - */ - - -#ifndef _RAR_REGISTER_H -#define _RAR_REGISTER_H - -#include - -/* following are used both in drivers as well as user space apps */ - -#define RAR_TYPE_VIDEO 0 -#define RAR_TYPE_AUDIO 1 -#define RAR_TYPE_IMAGE 2 -#define RAR_TYPE_DATA 3 - -#ifdef __KERNEL__ - -struct rar_device; - -int register_rar(int num, - int (*callback)(unsigned long data), unsigned long data); -void unregister_rar(int num); -int rar_get_address(int rar_index, dma_addr_t *start, dma_addr_t *end); -int rar_lock(int rar_index); - -#endif /* __KERNEL__ */ -#endif /* _RAR_REGISTER_H */ diff --git a/include/linux/rar_register.h b/include/linux/rar_register.h new file mode 100644 index 000000000000..ffa805780f85 --- /dev/null +++ b/include/linux/rar_register.h @@ -0,0 +1,44 @@ +/* + * Copyright (C) 2010 Intel Corporation. All rights reserved. + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of version 2 of the GNU General + * Public License as published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be + * useful, but WITHOUT ANY WARRANTY; without even the implied + * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR + * PURPOSE. See the GNU General Public License for more details. + * You should have received a copy of the GNU General Public + * License along with this program; if not, write to the Free + * Software Foundation, Inc., 59 Temple Place - Suite 330, + * Boston, MA 02111-1307, USA. + * The full GNU General Public License is included in this + * distribution in the file called COPYING. + */ + + +#ifndef _RAR_REGISTER_H +#define _RAR_REGISTER_H + +#include + +/* following are used both in drivers as well as user space apps */ + +#define RAR_TYPE_VIDEO 0 +#define RAR_TYPE_AUDIO 1 +#define RAR_TYPE_IMAGE 2 +#define RAR_TYPE_DATA 3 + +#ifdef __KERNEL__ + +struct rar_device; + +int register_rar(int num, + int (*callback)(unsigned long data), unsigned long data); +void unregister_rar(int num); +int rar_get_address(int rar_index, dma_addr_t *start, dma_addr_t *end); +int rar_lock(int rar_index); + +#endif /* __KERNEL__ */ +#endif /* _RAR_REGISTER_H */ -- cgit v1.2.3-70-g09d2 From 8950778704cf8483cc5cc0140f557adf0d3f45a5 Mon Sep 17 00:00:00 2001 From: Alek Du Date: Tue, 13 Jul 2010 10:56:25 +0100 Subject: gpio: Add PMIC GPIO block support Moorestown has PMIC chip which contains GPIO blocks. The PMIC chip is connected to Langwell by SPI interface. So this GPIO driver will be regarded as SPI GPIO expander though the actual GPIO access is through IPC and SRAM. The SPI master contoller will probe this device driver by parsing SPIB table. Cleaned up for new IPC, GPE removed and some printk and other tidying by Alan Cox. Fixes for points noted by Matthew Garrett Signed-off-by: Alek Du Signed-off-by: Alan Cox Signed-off-by: Matthew Garrett --- drivers/platform/x86/Kconfig | 7 + drivers/platform/x86/Makefile | 2 + drivers/platform/x86/intel_pmic_gpio.c | 340 +++++++++++++++++++++++++++++++++ include/linux/intel_pmic_gpio.h | 15 ++ 4 files changed, 364 insertions(+) create mode 100644 drivers/platform/x86/intel_pmic_gpio.c create mode 100644 include/linux/intel_pmic_gpio.h (limited to 'include') diff --git a/drivers/platform/x86/Kconfig b/drivers/platform/x86/Kconfig index 2189565d9e0e..ca30e561859e 100644 --- a/drivers/platform/x86/Kconfig +++ b/drivers/platform/x86/Kconfig @@ -540,6 +540,13 @@ config INTEL_SCU_IPC some embedded Intel x86 platforms. This is not needed for PC-type machines. +config GPIO_INTEL_PMIC + bool "Intel PMIC GPIO support" + depends on INTEL_SCU_IPC && GPIOLIB + ---help--- + Say Y here to support GPIO via the SCU IPC interface + on Intel MID platforms. + config RAR_REGISTER bool "Restricted Access Region Register Driver" depends on PCI && X86_MRST diff --git a/drivers/platform/x86/Makefile b/drivers/platform/x86/Makefile index ed50eca1b556..4744c7744ffa 100644 --- a/drivers/platform/x86/Makefile +++ b/drivers/platform/x86/Makefile @@ -28,3 +28,5 @@ obj-$(CONFIG_TOSHIBA_BT_RFKILL) += toshiba_bluetooth.o obj-$(CONFIG_INTEL_SCU_IPC) += intel_scu_ipc.o obj-$(CONFIG_RAR_REGISTER) += intel_rar_register.o obj-$(CONFIG_INTEL_IPS) += intel_ips.o +obj-$(CONFIG_GPIO_INTEL_PMIC) += intel_pmic_gpio.o + diff --git a/drivers/platform/x86/intel_pmic_gpio.c b/drivers/platform/x86/intel_pmic_gpio.c new file mode 100644 index 000000000000..5cdcff653918 --- /dev/null +++ b/drivers/platform/x86/intel_pmic_gpio.c @@ -0,0 +1,340 @@ +/* Moorestown PMIC GPIO (access through IPC) driver + * Copyright (c) 2008 - 2009, Intel Corporation. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + */ + +/* Supports: + * Moorestown platform PMIC chip + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define DRIVER_NAME "pmic_gpio" + +/* register offset that IPC driver should use + * 8 GPIO + 8 GPOSW (6 controllable) + 8GPO + */ +enum pmic_gpio_register { + GPIO0 = 0xE0, + GPIO7 = 0xE7, + GPIOINT = 0xE8, + GPOSWCTL0 = 0xEC, + GPOSWCTL5 = 0xF1, + GPO = 0xF4, +}; + +/* bits definition for GPIO & GPOSW */ +#define GPIO_DRV 0x01 +#define GPIO_DIR 0x02 +#define GPIO_DIN 0x04 +#define GPIO_DOU 0x08 +#define GPIO_INTCTL 0x30 +#define GPIO_DBC 0xc0 + +#define GPOSW_DRV 0x01 +#define GPOSW_DOU 0x08 +#define GPOSW_RDRV 0x30 + + +#define NUM_GPIO 24 + +struct pmic_gpio_irq { + spinlock_t lock; + u32 trigger[NUM_GPIO]; + u32 dirty; + struct work_struct work; +}; + + +struct pmic_gpio { + struct gpio_chip chip; + struct pmic_gpio_irq irqtypes; + void *gpiointr; + int irq; + unsigned irq_base; +}; + +static void pmic_program_irqtype(int gpio, int type) +{ + if (type & IRQ_TYPE_EDGE_RISING) + intel_scu_ipc_update_register(GPIO0 + gpio, 0x20, 0x20); + else + intel_scu_ipc_update_register(GPIO0 + gpio, 0x00, 0x20); + + if (type & IRQ_TYPE_EDGE_FALLING) + intel_scu_ipc_update_register(GPIO0 + gpio, 0x10, 0x10); + else + intel_scu_ipc_update_register(GPIO0 + gpio, 0x00, 0x10); +}; + +static void pmic_irqtype_work(struct work_struct *work) +{ + struct pmic_gpio_irq *t = + container_of(work, struct pmic_gpio_irq, work); + unsigned long flags; + int i; + u16 type; + + spin_lock_irqsave(&t->lock, flags); + /* As we drop the lock, we may need multiple scans if we race the + pmic_irq_type function */ + while (t->dirty) { + /* + * For each pin that has the dirty bit set send an IPC + * message to configure the hardware via the PMIC + */ + for (i = 0; i < NUM_GPIO; i++) { + if (!(t->dirty & (1 << i))) + continue; + t->dirty &= ~(1 << i); + /* We can't trust the array entry or dirty + once the lock is dropped */ + type = t->trigger[i]; + spin_unlock_irqrestore(&t->lock, flags); + pmic_program_irqtype(i, type); + spin_lock_irqsave(&t->lock, flags); + } + } + spin_unlock_irqrestore(&t->lock, flags); +} + +static int pmic_gpio_direction_input(struct gpio_chip *chip, unsigned offset) +{ + if (offset > 8) { + printk(KERN_ERR + "%s: only pin 0-7 support input\n", __func__); + return -1;/* we only have 8 GPIO can use as input */ + } + return intel_scu_ipc_update_register(GPIO0 + offset, + GPIO_DIR, GPIO_DIR); +} + +static int pmic_gpio_direction_output(struct gpio_chip *chip, + unsigned offset, int value) +{ + int rc = 0; + + if (offset < 8)/* it is GPIO */ + rc = intel_scu_ipc_update_register(GPIO0 + offset, + GPIO_DRV | GPIO_DOU | GPIO_DIR, + GPIO_DRV | (value ? GPIO_DOU : 0)); + else if (offset < 16)/* it is GPOSW */ + rc = intel_scu_ipc_update_register(GPOSWCTL0 + offset - 8, + GPOSW_DRV | GPOSW_DOU | GPOSW_RDRV, + GPOSW_DRV | (value ? GPOSW_DOU : 0)); + else if (offset > 15 && offset < 24)/* it is GPO */ + rc = intel_scu_ipc_update_register(GPO, + 1 << (offset - 16), + value ? 1 << (offset - 16) : 0); + else { + printk(KERN_ERR + "%s: invalid PMIC GPIO pin %d!\n", __func__, offset); + WARN_ON(1); + } + + return rc; +} + +static int pmic_gpio_get(struct gpio_chip *chip, unsigned offset) +{ + u8 r; + int ret; + + /* we only have 8 GPIO pins we can use as input */ + if (offset > 8) + return -EOPNOTSUPP; + ret = intel_scu_ipc_ioread8(GPIO0 + offset, &r); + if (ret < 0) + return ret; + return r & GPIO_DIN; +} + +static void pmic_gpio_set(struct gpio_chip *chip, unsigned offset, int value) +{ + if (offset < 8)/* it is GPIO */ + intel_scu_ipc_update_register(GPIO0 + offset, + GPIO_DRV | GPIO_DOU, + GPIO_DRV | (value ? GPIO_DOU : 0)); + else if (offset < 16)/* it is GPOSW */ + intel_scu_ipc_update_register(GPOSWCTL0 + offset - 8, + GPOSW_DRV | GPOSW_DOU | GPOSW_RDRV, + GPOSW_DRV | (value ? GPOSW_DOU : 0)); + else if (offset > 15 && offset < 24) /* it is GPO */ + intel_scu_ipc_update_register(GPO, + 1 << (offset - 16), + value ? 1 << (offset - 16) : 0); +} + +static int pmic_irq_type(unsigned irq, unsigned type) +{ + struct pmic_gpio *pg = get_irq_chip_data(irq); + u32 gpio = irq - pg->irq_base; + unsigned long flags; + + if (gpio > pg->chip.ngpio) + return -EINVAL; + + spin_lock_irqsave(&pg->irqtypes.lock, flags); + pg->irqtypes.trigger[gpio] = type; + pg->irqtypes.dirty |= (1 << gpio); + spin_unlock_irqrestore(&pg->irqtypes.lock, flags); + schedule_work(&pg->irqtypes.work); + return 0; +} + + + +static int pmic_gpio_to_irq(struct gpio_chip *chip, unsigned offset) +{ + struct pmic_gpio *pg = container_of(chip, struct pmic_gpio, chip); + + return pg->irq_base + offset; +} + +/* the gpiointr register is read-clear, so just do nothing. */ +static void pmic_irq_unmask(unsigned irq) +{ +}; + +static void pmic_irq_mask(unsigned irq) +{ +}; + +static struct irq_chip pmic_irqchip = { + .name = "PMIC-GPIO", + .mask = pmic_irq_mask, + .unmask = pmic_irq_unmask, + .set_type = pmic_irq_type, +}; + +static void pmic_irq_handler(unsigned irq, struct irq_desc *desc) +{ + struct pmic_gpio *pg = (struct pmic_gpio *)get_irq_data(irq); + u8 intsts = *((u8 *)pg->gpiointr + 4); + int gpio; + + for (gpio = 0; gpio < 8; gpio++) { + if (intsts & (1 << gpio)) { + pr_debug("pmic pin %d triggered\n", gpio); + generic_handle_irq(pg->irq_base + gpio); + } + } + desc->chip->eoi(irq); +} + +static int __devinit platform_pmic_gpio_probe(struct platform_device *pdev) +{ + struct device *dev = &pdev->dev; + int irq = platform_get_irq(pdev, 0); + struct intel_pmic_gpio_platform_data *pdata = dev->platform_data; + + struct pmic_gpio *pg; + int retval; + int i; + + if (irq < 0) { + dev_dbg(dev, "no IRQ line\n"); + return -EINVAL; + } + + if (!pdata || !pdata->gpio_base || !pdata->irq_base) { + dev_dbg(dev, "incorrect or missing platform data\n"); + return -EINVAL; + } + + pg = kzalloc(sizeof(*pg), GFP_KERNEL); + if (!pg) + return -ENOMEM; + + dev_set_drvdata(dev, pg); + + pg->irq = irq; + /* setting up SRAM mapping for GPIOINT register */ + pg->gpiointr = ioremap_nocache(pdata->gpiointr, 8); + if (!pg->gpiointr) { + printk(KERN_ERR "%s: Can not map GPIOINT.\n", __func__); + retval = -EINVAL; + goto err2; + } + pg->irq_base = pdata->irq_base; + pg->chip.label = "intel_pmic"; + pg->chip.direction_input = pmic_gpio_direction_input; + pg->chip.direction_output = pmic_gpio_direction_output; + pg->chip.get = pmic_gpio_get; + pg->chip.set = pmic_gpio_set; + pg->chip.to_irq = pmic_gpio_to_irq; + pg->chip.base = pdata->gpio_base; + pg->chip.ngpio = NUM_GPIO; + pg->chip.can_sleep = 1; + pg->chip.dev = dev; + + INIT_WORK(&pg->irqtypes.work, pmic_irqtype_work); + spin_lock_init(&pg->irqtypes.lock); + + pg->chip.dev = dev; + retval = gpiochip_add(&pg->chip); + if (retval) { + printk(KERN_ERR "%s: Can not add pmic gpio chip.\n", __func__); + goto err; + } + set_irq_data(pg->irq, pg); + set_irq_chained_handler(pg->irq, pmic_irq_handler); + for (i = 0; i < 8; i++) { + set_irq_chip_and_handler_name(i + pg->irq_base, &pmic_irqchip, + handle_simple_irq, "demux"); + set_irq_chip_data(i + pg->irq_base, pg); + } + return 0; +err: + iounmap(pg->gpiointr); +err2: + kfree(pg); + return retval; +} + +/* at the same time, register a platform driver + * this supports the sfi 0.81 fw */ +static struct platform_driver platform_pmic_gpio_driver = { + .driver = { + .name = DRIVER_NAME, + .owner = THIS_MODULE, + }, + .probe = platform_pmic_gpio_probe, +}; + +static int __init platform_pmic_gpio_init(void) +{ + return platform_driver_register(&platform_pmic_gpio_driver); +} + +subsys_initcall(platform_pmic_gpio_init); + +MODULE_AUTHOR("Alek Du "); +MODULE_DESCRIPTION("Intel Moorestown PMIC GPIO driver"); +MODULE_LICENSE("GPL v2"); diff --git a/include/linux/intel_pmic_gpio.h b/include/linux/intel_pmic_gpio.h new file mode 100644 index 000000000000..920109a29191 --- /dev/null +++ b/include/linux/intel_pmic_gpio.h @@ -0,0 +1,15 @@ +#ifndef LINUX_INTEL_PMIC_H +#define LINUX_INTEL_PMIC_H + +struct intel_pmic_gpio_platform_data { + /* the first IRQ of the chip */ + unsigned irq_base; + /* number assigned to the first GPIO */ + unsigned gpio_base; + /* sram address for gpiointr register, the langwell chip will map + * the PMIC spi GPIO expander's GPIOINTR register in sram. + */ + unsigned gpiointr; +}; + +#endif -- cgit v1.2.3-70-g09d2